kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-14 06:27:45 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity
42d0585493
TensorRT-LLMs/_modules/tensorrt_llm/models/mllama/model.html
nv-guomingz 42d0585493
doc:fix llm api reference blank issue. (#2669)
2025-01-08 14:45:46 +08:00

1732 lines
234 KiB
HTML
Raw Blame History

<!DOCTYPE html>
<html class="writer-html5" lang="en" data-content_root="../../../../">
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>tensorrt_llm.models.mllama.model &mdash; tensorrt_llm documentation</title>
<link rel="stylesheet" type="text/css" href="../../../../_static/pygments.css?v=80d5e7a1" />
<link rel="stylesheet" type="text/css" href="../../../../_static/css/theme.css?v=e59714d7" />
<link rel="stylesheet" type="text/css" href="../../../../_static/copybutton.css?v=76b2166b" />
<script src="../../../../_static/jquery.js?v=5d32c60e"></script>
<script src="../../../../_static/_sphinx_javascript_frameworks_compat.js?v=2cd50e6c"></script>
<script src="../../../../_static/documentation_options.js?v=5929fcd5"></script>
<script src="../../../../_static/doctools.js?v=9bcbadda"></script>
<script src="../../../../_static/sphinx_highlight.js?v=dc90522c"></script>
<script src="../../../../_static/clipboard.min.js?v=a7894cd8"></script>
<script src="../../../../_static/copybutton.js?v=65e89d2a"></script>
<script src="../../../../_static/js/theme.js"></script>
<link rel="index" title="Index" href="../../../../genindex.html" />
<link rel="search" title="Search" href="../../../../search.html" />
</head>
<body class="wy-body-for-nav">
<div class="wy-grid-for-nav">
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
<div class="wy-side-scroll">
<div class="wy-side-nav-search" >
<a href="../../../../index.html" class="icon icon-home">
tensorrt_llm
</a>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="../../../../search.html" method="get">
<input type="text" name="q" placeholder="Search docs" aria-label="Search docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div><div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="Navigation menu">
<p class="caption" role="heading"><span class="caption-text">Getting Started</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../../../overview.html">Overview</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../quick-start-guide.html">Quick Start Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../key-features.html">Key Features</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../release-notes.html">Release Notes</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Installation</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../../../installation/linux.html">Installing on Linux</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../installation/build-from-source-linux.html">Building from Source Code on Linux</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../installation/windows.html">Installing on Windows</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../installation/build-from-source-windows.html">Building from Source Code on Windows</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../installation/grace-hopper.html">Installing on Grace Hopper</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">LLM API</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../../../llm-api/index.html">API Introduction</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../llm-api/reference.html">API Reference</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">LLM API Examples</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../../../llm-api-examples/index.html">LLM Examples Introduction</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../llm-api-examples/customization.html">Common Customizations</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../llm-api-examples/llm_api_examples.html">Examples</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Model Definition API</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../../../python-api/tensorrt_llm.layers.html">Layers</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../python-api/tensorrt_llm.functional.html">Functionals</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../python-api/tensorrt_llm.models.html">Models</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../python-api/tensorrt_llm.plugin.html">Plugin</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../python-api/tensorrt_llm.quantization.html">Quantization</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../python-api/tensorrt_llm.runtime.html">Runtime</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">C++ API</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../../../_cpp_gen/executor.html">Executor</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../_cpp_gen/runtime.html">Runtime</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Command-Line Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../../../commands/trtllm-build.html">trtllm-build</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../commands/trtllm-serve.html">trtllm-serve</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Architecture</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../../../architecture/overview.html">TensorRT-LLM Architecture</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../architecture/core-concepts.html">Model Definition</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../architecture/core-concepts.html#compilation">Compilation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../architecture/core-concepts.html#runtime">Runtime</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../architecture/core-concepts.html#multi-gpu-and-multi-node-support">Multi-GPU and Multi-Node Support</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../architecture/checkpoint.html">TensorRT-LLM Checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../architecture/workflow.html">TensorRT-LLM Build Workflow</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../architecture/add-model.html">Adding a Model</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Advanced</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/gpt-attention.html">Multi-Head, Multi-Query, and Group-Query Attention</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/gpt-runtime.html">C++ GPT Runtime</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/executor.html">Executor API</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/graph-rewriting.html">Graph Rewriting Module</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/inference-request.html">Inference Request</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/inference-request.html#responses">Responses</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/lora.html">Run gpt-2b + LoRA using GptManager / cpp runtime</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/expert-parallelism.html">Expert Parallelism in TensorRT-LLM</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/kv-cache-reuse.html">KV cache reuse</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/speculative-decoding.html">Speculative Sampling</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../advanced/disaggregated-service.html">Disaggregated-Service (experimental)</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Performance</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../../../performance/perf-overview.html">Overview</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../performance/perf-benchmarking.html">Benchmarking</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../performance/perf-best-practices.html">Best Practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../performance/perf-analysis.html">Performance Analysis</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../../../reference/troubleshooting.html">Troubleshooting</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../reference/support-matrix.html">Support Matrix</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../reference/precision.html">Numerical Precision</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../reference/memory.html">Memory Usage of TensorRT-LLM</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Blogs</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/H100vsA100.html">H100 has 4.6x A100 Performance in TensorRT-LLM, achieving 10,000 tok/s at 100ms to first token</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/H200launch.html">H200 achieves nearly 12,000 tokens/sec on Llama2-13B with TensorRT-LLM</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/Falcon180B-H200.html">Falcon-180B on a single H200 GPU with INT4 AWQ, and 6.7x faster Llama-70B over A100</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/quantization-in-TRT-LLM.html">Speed up inference with SOTA quantization techniques in TRT-LLM</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../../blogs/XQA-kernel.html">New XQA-kernel provides 2.4x more Llama-70B throughput within the same latency budget</a></li>
</ul>
</div>
</div>
</nav>
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="Mobile navigation menu" >
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
<a href="../../../../index.html">tensorrt_llm</a>
</nav>
<div class="wy-nav-content">
<div class="rst-content">
<div role="navigation" aria-label="Page navigation">
<ul class="wy-breadcrumbs">
<li><a href="../../../../index.html" class="icon icon-home" aria-label="Home"></a></li>
<li class="breadcrumb-item"><a href="../../../index.html">Module code</a></li>
<li class="breadcrumb-item active">tensorrt_llm.models.mllama.model</li>
<li class="wy-breadcrumbs-aside">
</li>
</ul>
<hr/>
</div>
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
<div itemprop="articleBody">
<h1>Source code for tensorrt_llm.models.mllama.model</h1><div class="highlight"><pre>
<span></span><span class="c1"># SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION &amp; AFFILIATES. All rights reserved.</span>
<span class="c1"># SPDX-License-Identifier: Apache-2.0</span>
<span class="c1">#</span>
<span class="c1"># Licensed under the Apache License, Version 2.0 (the &quot;License&quot;);</span>
<span class="c1"># you may not use this file except in compliance with the License.</span>
<span class="c1"># You may obtain a copy of the License at</span>
<span class="c1">#</span>
<span class="c1"># http://www.apache.org/licenses/LICENSE-2.0</span>
<span class="c1">#</span>
<span class="c1"># Unless required by applicable law or agreed to in writing, software</span>
<span class="c1"># distributed under the License is distributed on an &quot;AS IS&quot; BASIS,</span>
<span class="c1"># WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.</span>
<span class="c1"># See the License for the specific language governing permissions and</span>
<span class="c1"># limitations under the License.</span>
<span class="kn">import</span> <span class="nn">math</span>
<span class="kn">from</span> <span class="nn">collections</span> <span class="kn">import</span> <span class="n">OrderedDict</span>
<span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">List</span><span class="p">,</span> <span class="n">Optional</span><span class="p">,</span> <span class="n">Union</span>
<span class="kn">import</span> <span class="nn">tensorrt</span> <span class="k">as</span> <span class="nn">trt</span>
<span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm._common</span> <span class="kn">import</span> <span class="n">default_net</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm._utils</span> <span class="kn">import</span> <span class="n">numpy_to_torch</span><span class="p">,</span> <span class="n">str_dtype_to_torch</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm.bindings</span> <span class="kn">import</span> <span class="n">KVCacheType</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm.functional</span> <span class="kn">import</span> <span class="p">(</span><span class="n">Conditional</span><span class="p">,</span> <span class="n">LayerNormPositionType</span><span class="p">,</span>
<span class="n">LayerNormType</span><span class="p">,</span> <span class="n">MLPType</span><span class="p">,</span>
<span class="n">PositionEmbeddingType</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">,</span> <span class="n">assertion</span><span class="p">,</span>
<span class="n">gather_last_token_logits</span><span class="p">,</span> <span class="n">maximum</span><span class="p">,</span> <span class="n">minimum</span><span class="p">,</span>
<span class="n">recv</span><span class="p">,</span> <span class="n">reduce</span><span class="p">,</span> <span class="n">send</span><span class="p">,</span> <span class="n">shape</span><span class="p">,</span> <span class="n">tanh</span><span class="p">)</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm.layers</span> <span class="kn">import</span> <span class="p">(</span><span class="n">MLP</span><span class="p">,</span> <span class="n">Attention</span><span class="p">,</span> <span class="n">AttentionMaskParams</span><span class="p">,</span>
<span class="n">AttentionMaskType</span><span class="p">,</span> <span class="n">AttentionParams</span><span class="p">,</span>
<span class="n">ColumnLinear</span><span class="p">,</span> <span class="n">Embedding</span><span class="p">,</span> <span class="n">FusedGatedMLP</span><span class="p">,</span>
<span class="n">GatedMLP</span><span class="p">,</span> <span class="n">GroupNorm</span><span class="p">,</span> <span class="n">KeyValueCacheParams</span><span class="p">,</span>
<span class="n">LayerNorm</span><span class="p">,</span> <span class="n">LoraParams</span><span class="p">,</span> <span class="n">RmsNorm</span><span class="p">)</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm.lora_manager</span> <span class="kn">import</span> <span class="p">(</span><span class="n">LoraConfig</span><span class="p">,</span>
<span class="n">get_default_trtllm_modules_to_hf_modules</span><span class="p">,</span>
<span class="n">use_lora</span><span class="p">)</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm.mapping</span> <span class="kn">import</span> <span class="n">Mapping</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm.models.model_weights_loader</span> <span class="kn">import</span> <span class="n">ModelWeightsLoader</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm.models.modeling_utils</span> <span class="kn">import</span> <span class="n">PretrainedModel</span><span class="p">,</span> <span class="n">QuantConfig</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm.module</span> <span class="kn">import</span> <span class="n">Module</span><span class="p">,</span> <span class="n">ModuleList</span>
<span class="kn">from</span> <span class="nn">tensorrt_llm.parameter</span> <span class="kn">import</span> <span class="n">Parameter</span>
<span class="kn">from</span> <span class="nn">.config</span> <span class="kn">import</span> <span class="n">MLLaMAConfig</span>
<span class="n">layernorm_map</span> <span class="o">=</span> <span class="p">{</span>
<span class="n">LayerNormType</span><span class="o">.</span><span class="n">LayerNorm</span><span class="p">:</span> <span class="n">LayerNorm</span><span class="p">,</span>
<span class="n">LayerNormType</span><span class="o">.</span><span class="n">RmsNorm</span><span class="p">:</span> <span class="n">RmsNorm</span><span class="p">,</span>
<span class="n">LayerNormType</span><span class="o">.</span><span class="n">GroupNorm</span><span class="p">:</span> <span class="n">GroupNorm</span><span class="p">,</span>
<span class="p">}</span>
<span class="n">mlp_map</span> <span class="o">=</span> <span class="p">{</span>
<span class="n">MLPType</span><span class="o">.</span><span class="n">MLP</span><span class="p">:</span> <span class="n">MLP</span><span class="p">,</span>
<span class="n">MLPType</span><span class="o">.</span><span class="n">GatedMLP</span><span class="p">:</span> <span class="n">GatedMLP</span><span class="p">,</span>
<span class="n">MLPType</span><span class="o">.</span><span class="n">FusedGatedMLP</span><span class="p">:</span> <span class="n">FusedGatedMLP</span><span class="p">,</span>
<span class="p">}</span>
<span class="n">ADD_DEBUG_TENSOR</span> <span class="o">=</span> <span class="kc">False</span>
<span class="k">class</span> <span class="nc">CrossAttentionTransformerBlock</span><span class="p">(</span><span class="n">Module</span><span class="p">):</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span>
<span class="bp">self</span><span class="p">,</span>
<span class="o">*</span><span class="p">,</span>
<span class="n">local_layer_idx</span><span class="p">,</span>
<span class="n">hidden_size</span><span class="p">,</span>
<span class="n">ffn_hidden_size</span><span class="p">,</span>
<span class="n">num_attention_heads</span><span class="p">,</span>
<span class="n">num_kv_heads</span><span class="p">,</span>
<span class="n">head_size</span><span class="p">,</span>
<span class="n">max_position_embeddings</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">q_scaling</span><span class="o">=</span><span class="mf">1.0</span><span class="p">,</span>
<span class="n">has_attention_qkvo_bias</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">has_mlp_bias</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">layernorm_position</span><span class="o">=</span><span class="n">LayerNormPositionType</span><span class="o">.</span><span class="n">pre_layernorm</span><span class="p">,</span>
<span class="n">layernorm_type</span><span class="o">=</span><span class="n">LayerNormType</span><span class="o">.</span><span class="n">RmsNorm</span><span class="p">,</span>
<span class="n">layernorm_eps</span><span class="o">=</span><span class="mf">1e-5</span><span class="p">,</span>
<span class="n">hidden_act</span><span class="o">=</span><span class="s2">&quot;gated-silu&quot;</span><span class="p">,</span>
<span class="n">mlp_type</span><span class="o">=</span><span class="n">MLPType</span><span class="o">.</span><span class="n">GatedMLP</span><span class="p">,</span>
<span class="n">mapping</span><span class="o">=</span><span class="n">Mapping</span><span class="p">(),</span>
<span class="n">dtype</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">residual_scaling</span><span class="o">=</span><span class="mf">1.0</span><span class="p">,</span>
<span class="n">relative_attention</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">max_distance</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span>
<span class="n">num_buckets</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span>
<span class="n">fp16_clamping</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">skip_cross_kv</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">use_implicit_relative_attention</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">rotary_embedding_base</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">rotary_embedding_scaling</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">layer_idx_in_cache_pool</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="p">):</span>
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span> <span class="o">=</span> <span class="n">local_layer_idx</span>
<span class="bp">self</span><span class="o">.</span><span class="n">layernorm_type</span> <span class="o">=</span> <span class="n">layernorm_type</span>
<span class="n">ln_type</span> <span class="o">=</span> <span class="n">layernorm_map</span><span class="p">[</span><span class="n">layernorm_type</span><span class="p">]</span>
<span class="bp">self</span><span class="o">.</span><span class="n">layernorm_position</span> <span class="o">=</span> <span class="n">layernorm_position</span>
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">layernorm_position</span> <span class="o">==</span> <span class="n">LayerNormPositionType</span><span class="o">.</span><span class="n">pre_layernorm</span>
<span class="bp">self</span><span class="o">.</span><span class="n">cross_attention</span> <span class="o">=</span> <span class="n">Attention</span><span class="p">(</span>
<span class="n">local_layer_idx</span><span class="o">=</span><span class="n">local_layer_idx</span><span class="p">,</span>
<span class="n">hidden_size</span><span class="o">=</span><span class="n">hidden_size</span><span class="p">,</span>
<span class="n">num_attention_heads</span><span class="o">=</span><span class="n">num_attention_heads</span><span class="p">,</span>
<span class="n">attention_head_size</span><span class="o">=</span><span class="n">head_size</span><span class="p">,</span>
<span class="n">num_kv_heads</span><span class="o">=</span><span class="n">num_kv_heads</span><span class="p">,</span>
<span class="n">max_position_embeddings</span><span class="o">=</span><span class="n">max_position_embeddings</span><span class="p">,</span>
<span class="n">q_scaling</span><span class="o">=</span><span class="n">q_scaling</span><span class="p">,</span>
<span class="n">bias</span><span class="o">=</span><span class="n">has_attention_qkvo_bias</span><span class="p">,</span>
<span class="n">attention_mask_type</span><span class="o">=</span><span class="n">AttentionMaskType</span><span class="o">.</span><span class="n">causal</span><span class="p">,</span>
<span class="n">tp_group</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
<span class="n">tp_size</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="n">tp_rank</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_rank</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span>
<span class="n">cross_attention</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">relative_attention</span><span class="o">=</span>
<span class="kc">False</span><span class="p">,</span> <span class="c1"># Cross attention has no relative attention bias</span>
<span class="n">max_distance</span><span class="o">=</span><span class="n">max_distance</span><span class="p">,</span>
<span class="n">num_buckets</span><span class="o">=</span><span class="n">num_buckets</span><span class="p">,</span>
<span class="n">position_embedding_type</span><span class="o">=</span><span class="n">PositionEmbeddingType</span><span class="o">.</span>
<span class="n">learned_absolute</span><span class="p">,</span> <span class="c1"># we don&#39;t use rope for cross attn</span>
<span class="n">skip_cross_kv</span><span class="o">=</span><span class="n">skip_cross_kv</span><span class="p">,</span>
<span class="n">qk_layernorm</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">layernorm_type</span><span class="o">=</span><span class="n">layernorm_type</span><span class="p">,</span>
<span class="n">layer_idx_in_cache_pool</span><span class="o">=</span><span class="n">layer_idx_in_cache_pool</span><span class="p">,</span>
<span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">input_layernorm</span> <span class="o">=</span> <span class="n">ln_type</span><span class="p">(</span><span class="n">normalized_shape</span><span class="o">=</span><span class="n">hidden_size</span><span class="p">,</span>
<span class="n">eps</span><span class="o">=</span><span class="n">layernorm_eps</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">gate_attn</span> <span class="o">=</span> <span class="n">Parameter</span><span class="p">(</span><span class="n">shape</span><span class="o">=</span><span class="nb">tuple</span><span class="p">((</span><span class="mi">1</span><span class="p">,</span> <span class="p">)),</span> <span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mlp_type</span> <span class="o">=</span> <span class="n">mlp_type</span>
<span class="n">mlp_f</span> <span class="o">=</span> <span class="n">mlp_map</span><span class="p">[</span><span class="n">mlp_type</span><span class="p">]</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mlp</span> <span class="o">=</span> <span class="n">mlp_f</span><span class="p">(</span>
<span class="n">hidden_size</span><span class="o">=</span><span class="n">hidden_size</span><span class="p">,</span>
<span class="n">ffn_hidden_size</span><span class="o">=</span><span class="n">ffn_hidden_size</span><span class="p">,</span>
<span class="n">hidden_act</span><span class="o">=</span><span class="n">hidden_act</span><span class="p">,</span>
<span class="n">bias</span><span class="o">=</span><span class="n">has_mlp_bias</span><span class="p">,</span>
<span class="n">tp_group</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
<span class="n">tp_size</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span>
<span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">post_layernorm</span> <span class="o">=</span> <span class="n">ln_type</span><span class="p">(</span><span class="n">normalized_shape</span><span class="o">=</span><span class="n">hidden_size</span><span class="p">,</span>
<span class="n">eps</span><span class="o">=</span><span class="n">layernorm_eps</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">gate_ffwd</span> <span class="o">=</span> <span class="n">Parameter</span><span class="p">(</span><span class="n">shape</span><span class="o">=</span><span class="nb">tuple</span><span class="p">((</span><span class="mi">1</span><span class="p">,</span> <span class="p">)),</span> <span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">residual_scaling</span> <span class="o">=</span> <span class="n">residual_scaling</span>
<span class="bp">self</span><span class="o">.</span><span class="n">fp16_clamping</span> <span class="o">=</span> <span class="n">fp16_clamping</span>
<span class="bp">self</span><span class="o">.</span><span class="n">no_ffn</span> <span class="o">=</span> <span class="kc">False</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
<span class="n">encoder_output</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">attention_mask_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">use_cache</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">kv_cache_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">attention_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">lora_layer_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">cross_kv_cache_gen</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">cross_kv_reuse</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">full_text_row_masked_out_mask</span><span class="p">:</span> <span class="n">Tensor</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">skip_cross_attn_blocks</span><span class="p">:</span> <span class="n">Tensor</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span>
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">)</span>
<span class="k">if</span> <span class="n">encoder_output</span><span class="p">:</span>
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">encoder_output</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">)</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/1.0: hidden_states&#39;</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="c1"># cross attention</span>
<span class="n">residual</span> <span class="o">=</span> <span class="n">hidden_states</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">residual_scaling</span>
<span class="c1"># skip input_layernorm</span>
<span class="k">if</span> <span class="n">skip_cross_attn_blocks</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">input_ln_conditional</span> <span class="o">=</span> <span class="n">Conditional</span><span class="p">(</span><span class="n">skip_cross_attn_blocks</span><span class="p">)</span>
<span class="n">skip_result</span> <span class="o">=</span> <span class="n">input_ln_conditional</span><span class="o">.</span><span class="n">add_input</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">input_ln_conditional</span><span class="o">.</span><span class="n">add_input</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">input_layernorm</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">input_ln_conditional</span><span class="o">.</span><span class="n">add_output</span><span class="p">(</span>
<span class="n">skip_result</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">input_layernorm</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/2.1: normed_input&#39;</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="c1"># pass full_text_row_masked_out_mask and xattn_mask</span>
<span class="n">attention_output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">cross_attention</span><span class="p">(</span>
<span class="n">hidden_states</span><span class="o">=</span><span class="n">hidden_states</span><span class="p">,</span>
<span class="n">attention_mask</span><span class="o">=</span><span class="n">attention_mask_params</span><span class="o">.</span><span class="n">cross_attention_mask</span><span class="p">,</span>
<span class="n">attention_packed_mask</span><span class="o">=</span><span class="n">attention_mask_params</span><span class="o">.</span>
<span class="n">cross_attention_packed_mask</span><span class="p">,</span>
<span class="n">encoder_output</span><span class="o">=</span><span class="n">encoder_output</span><span class="p">,</span>
<span class="n">use_cache</span><span class="o">=</span><span class="n">use_cache</span><span class="p">,</span>
<span class="n">kv_cache_params</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="p">,</span>
<span class="n">attention_params</span><span class="o">=</span><span class="n">attention_params</span><span class="p">,</span>
<span class="n">lora_layer_params</span><span class="o">=</span><span class="n">lora_layer_params</span><span class="p">,</span>
<span class="n">cross_kv_cache_gen</span><span class="o">=</span><span class="n">cross_kv_cache_gen</span><span class="p">,</span>
<span class="n">cross_kv_reuse</span><span class="o">=</span><span class="n">cross_kv_reuse</span><span class="p">,</span>
<span class="n">skip_attn</span><span class="o">=</span><span class="n">skip_cross_attn_blocks</span><span class="p">,</span>
<span class="p">)</span>
<span class="k">if</span> <span class="n">use_cache</span><span class="p">:</span>
<span class="n">attention_output</span><span class="p">,</span> <span class="n">presents_cross</span> <span class="o">=</span> <span class="n">attention_output</span>
<span class="n">attention_output</span> <span class="o">=</span> <span class="n">attention_output</span> <span class="o">*</span> <span class="n">full_text_row_masked_out_mask</span> <span class="c1"># TODO(bhsueh) should move this mask into attention?</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">attention_output</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/3.1: cross_attention_output&#39;</span><span class="p">,</span>
<span class="n">attention_output</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="n">attn_residual_scale</span> <span class="o">=</span> <span class="n">tanh</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">gate_attn</span><span class="o">.</span><span class="n">value</span><span class="o">.</span><span class="n">cast</span><span class="p">(</span><span class="n">trt</span><span class="o">.</span><span class="n">float32</span><span class="p">))</span><span class="o">.</span><span class="n">cast</span><span class="p">(</span>
<span class="n">attention_output</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="n">attention_input</span> <span class="o">=</span> <span class="n">hidden_states</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">residual</span> <span class="o">+</span> <span class="n">attn_residual_scale</span> <span class="o">*</span> <span class="n">attention_output</span>
<span class="c1"># use to skip attention_output with residual</span>
<span class="c1"># Since conditional does not work for gpt_attention_plugin, we replace the</span>
<span class="c1"># attention_output by hidden_states (input of attention) now.</span>
<span class="k">if</span> <span class="n">skip_cross_attn_blocks</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">attn_conditional</span> <span class="o">=</span> <span class="n">Conditional</span><span class="p">(</span><span class="n">skip_cross_attn_blocks</span><span class="p">)</span>
<span class="n">skip_result</span> <span class="o">=</span> <span class="n">attn_conditional</span><span class="o">.</span><span class="n">add_input</span><span class="p">(</span><span class="n">attention_input</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">attn_conditional</span><span class="o">.</span><span class="n">add_input</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">attn_conditional</span><span class="o">.</span><span class="n">add_output</span><span class="p">(</span><span class="n">skip_result</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="p">)</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/3.2: cross_attn_output_with_residual&#39;</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">fp16_clamping</span><span class="p">:</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">maximum</span><span class="p">(</span><span class="o">-</span><span class="mf">64000.0</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">minimum</span><span class="p">(</span><span class="mf">64000.0</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">)</span>
<span class="c1"># MLP</span>
<span class="c1"># skip post_layernorm and mlp</span>
<span class="k">if</span> <span class="n">skip_cross_attn_blocks</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">mlp_conditional</span> <span class="o">=</span> <span class="n">Conditional</span><span class="p">(</span><span class="n">skip_cross_attn_blocks</span><span class="p">)</span>
<span class="n">skip_case</span> <span class="o">=</span> <span class="n">mlp_conditional</span><span class="o">.</span><span class="n">add_input</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">mlp_conditional</span><span class="o">.</span><span class="n">add_input</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="n">residual</span> <span class="o">=</span> <span class="n">hidden_states</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">residual_scaling</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">post_layernorm</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">mlp</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">,</span>
<span class="n">lora_layer_params</span><span class="o">=</span><span class="n">lora_layer_params</span><span class="p">)</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/4.1: mlp_output&#39;</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">hidden_states</span> <span class="o">*</span> <span class="n">full_text_row_masked_out_mask</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/4.2: masked_mlp_output&#39;</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="n">ffn_residual_scale</span> <span class="o">=</span> <span class="n">tanh</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">gate_ffwd</span><span class="o">.</span><span class="n">value</span><span class="o">.</span><span class="n">cast</span><span class="p">(</span><span class="n">trt</span><span class="o">.</span><span class="n">float32</span><span class="p">))</span><span class="o">.</span><span class="n">cast</span><span class="p">(</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">residual</span> <span class="o">+</span> <span class="n">ffn_residual_scale</span> <span class="o">*</span> <span class="n">hidden_states</span> <span class="o">*</span> <span class="nb">float</span><span class="p">(</span>
<span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">no_ffn</span><span class="p">)</span>
<span class="k">if</span> <span class="n">skip_cross_attn_blocks</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">mlp_conditional</span><span class="o">.</span><span class="n">add_output</span><span class="p">(</span><span class="n">skip_case</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">)</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">fp16_clamping</span><span class="p">:</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">maximum</span><span class="p">(</span><span class="o">-</span><span class="mf">64000.0</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">minimum</span><span class="p">(</span><span class="mf">64000.0</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">)</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/4.4: transformer_out&#39;</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="k">if</span> <span class="n">use_cache</span><span class="p">:</span>
<span class="k">return</span> <span class="p">(</span><span class="n">hidden_states</span><span class="p">,</span> <span class="n">presents_cross</span><span class="p">)</span>
<span class="k">return</span> <span class="n">hidden_states</span>
<span class="k">class</span> <span class="nc">TransformerBlock</span><span class="p">(</span><span class="n">Module</span><span class="p">):</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span>
<span class="bp">self</span><span class="p">,</span>
<span class="o">*</span><span class="p">,</span>
<span class="n">local_layer_idx</span><span class="p">,</span>
<span class="n">hidden_size</span><span class="p">,</span>
<span class="n">ffn_hidden_size</span><span class="p">,</span>
<span class="n">num_attention_heads</span><span class="p">,</span>
<span class="n">num_kv_heads</span><span class="p">,</span>
<span class="n">head_size</span><span class="p">,</span>
<span class="n">max_position_embeddings</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">q_scaling</span><span class="o">=</span><span class="mf">1.0</span><span class="p">,</span>
<span class="n">has_attention_qkvo_bias</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">has_mlp_bias</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">layernorm_position</span><span class="o">=</span><span class="n">LayerNormPositionType</span><span class="o">.</span><span class="n">pre_layernorm</span><span class="p">,</span>
<span class="n">layernorm_type</span><span class="o">=</span><span class="n">LayerNormType</span><span class="o">.</span><span class="n">RmsNorm</span><span class="p">,</span>
<span class="n">layernorm_eps</span><span class="o">=</span><span class="mf">1e-5</span><span class="p">,</span>
<span class="n">hidden_act</span><span class="o">=</span><span class="s2">&quot;gated-silu&quot;</span><span class="p">,</span>
<span class="n">mlp_type</span><span class="o">=</span><span class="n">MLPType</span><span class="o">.</span><span class="n">GatedMLP</span><span class="p">,</span>
<span class="n">mapping</span><span class="o">=</span><span class="n">Mapping</span><span class="p">(),</span>
<span class="n">dtype</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">residual_scaling</span><span class="o">=</span><span class="mf">1.0</span><span class="p">,</span>
<span class="n">relative_attention</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">max_distance</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span>
<span class="n">num_buckets</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span>
<span class="n">fp16_clamping</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">skip_cross_kv</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">use_implicit_relative_attention</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">rotary_embedding_base</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">rotary_embedding_scaling</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">layer_idx_in_cache_pool</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="p">):</span>
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span> <span class="o">=</span> <span class="n">local_layer_idx</span>
<span class="bp">self</span><span class="o">.</span><span class="n">layernorm_type</span> <span class="o">=</span> <span class="n">layernorm_type</span>
<span class="n">ln_type</span> <span class="o">=</span> <span class="n">layernorm_map</span><span class="p">[</span><span class="n">layernorm_type</span><span class="p">]</span>
<span class="bp">self</span><span class="o">.</span><span class="n">layernorm_position</span> <span class="o">=</span> <span class="n">layernorm_position</span>
<span class="k">assert</span> <span class="bp">self</span><span class="o">.</span><span class="n">layernorm_position</span> <span class="o">==</span> <span class="n">LayerNormPositionType</span><span class="o">.</span><span class="n">pre_layernorm</span>
<span class="bp">self</span><span class="o">.</span><span class="n">self_attention</span> <span class="o">=</span> <span class="n">Attention</span><span class="p">(</span>
<span class="n">local_layer_idx</span><span class="o">=</span><span class="n">local_layer_idx</span><span class="p">,</span>
<span class="n">hidden_size</span><span class="o">=</span><span class="n">hidden_size</span><span class="p">,</span>
<span class="n">num_attention_heads</span><span class="o">=</span><span class="n">num_attention_heads</span><span class="p">,</span>
<span class="n">attention_head_size</span><span class="o">=</span><span class="n">head_size</span><span class="p">,</span>
<span class="n">num_kv_heads</span><span class="o">=</span><span class="n">num_kv_heads</span><span class="p">,</span>
<span class="n">max_position_embeddings</span><span class="o">=</span><span class="n">max_position_embeddings</span><span class="p">,</span>
<span class="n">q_scaling</span><span class="o">=</span><span class="n">q_scaling</span><span class="p">,</span>
<span class="n">bias</span><span class="o">=</span><span class="n">has_attention_qkvo_bias</span><span class="p">,</span>
<span class="n">attention_mask_type</span><span class="o">=</span><span class="n">AttentionMaskType</span><span class="o">.</span><span class="n">causal</span><span class="p">,</span>
<span class="n">tp_group</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
<span class="n">tp_size</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="n">tp_rank</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_rank</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span>
<span class="n">cross_attention</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">relative_attention</span><span class="o">=</span><span class="n">relative_attention</span><span class="p">,</span>
<span class="n">max_distance</span><span class="o">=</span><span class="n">max_distance</span> <span class="k">if</span> <span class="n">use_implicit_relative_attention</span> <span class="k">else</span> <span class="mi">0</span><span class="p">,</span>
<span class="n">num_buckets</span><span class="o">=</span><span class="n">num_buckets</span><span class="p">,</span>
<span class="n">position_embedding_type</span><span class="o">=</span><span class="n">PositionEmbeddingType</span><span class="o">.</span><span class="n">relative</span>
<span class="k">if</span> <span class="n">relative_attention</span> <span class="k">else</span> <span class="n">PositionEmbeddingType</span><span class="o">.</span><span class="n">rope_gpt_neox</span><span class="p">,</span>
<span class="n">use_implicit_relative_attention</span><span class="o">=</span><span class="n">use_implicit_relative_attention</span><span class="p">,</span>
<span class="n">rotary_embedding_base</span><span class="o">=</span><span class="n">rotary_embedding_base</span><span class="p">,</span>
<span class="n">rotary_embedding_scaling</span><span class="o">=</span><span class="n">rotary_embedding_scaling</span><span class="p">,</span>
<span class="n">layer_idx_in_cache_pool</span><span class="o">=</span><span class="n">layer_idx_in_cache_pool</span><span class="p">,</span>
<span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">input_layernorm</span> <span class="o">=</span> <span class="n">ln_type</span><span class="p">(</span><span class="n">normalized_shape</span><span class="o">=</span><span class="n">hidden_size</span><span class="p">,</span>
<span class="n">eps</span><span class="o">=</span><span class="n">layernorm_eps</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mlp_type</span> <span class="o">=</span> <span class="n">mlp_type</span>
<span class="n">mlp_f</span> <span class="o">=</span> <span class="n">mlp_map</span><span class="p">[</span><span class="n">mlp_type</span><span class="p">]</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mlp</span> <span class="o">=</span> <span class="n">mlp_f</span><span class="p">(</span>
<span class="n">hidden_size</span><span class="o">=</span><span class="n">hidden_size</span><span class="p">,</span>
<span class="n">ffn_hidden_size</span><span class="o">=</span><span class="n">ffn_hidden_size</span><span class="p">,</span>
<span class="n">hidden_act</span><span class="o">=</span><span class="n">hidden_act</span><span class="p">,</span>
<span class="n">bias</span><span class="o">=</span><span class="n">has_mlp_bias</span><span class="p">,</span>
<span class="n">tp_group</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
<span class="n">tp_size</span><span class="o">=</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span>
<span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">post_layernorm</span> <span class="o">=</span> <span class="n">ln_type</span><span class="p">(</span><span class="n">normalized_shape</span><span class="o">=</span><span class="n">hidden_size</span><span class="p">,</span>
<span class="n">eps</span><span class="o">=</span><span class="n">layernorm_eps</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">residual_scaling</span> <span class="o">=</span> <span class="n">residual_scaling</span>
<span class="bp">self</span><span class="o">.</span><span class="n">fp16_clamping</span> <span class="o">=</span> <span class="n">fp16_clamping</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span>
<span class="bp">self</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
<span class="n">encoder_output</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="c1"># not used</span>
<span class="n">attention_mask_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">use_cache</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">kv_cache_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">attention_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">lora_layer_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">cross_kv_cache_gen</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">cross_kv_reuse</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">full_text_row_masked_out_mask</span><span class="p">:</span> <span class="n">Tensor</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="c1"># not used</span>
<span class="n">skip_cross_attn_blocks</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="p">):</span>
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">)</span>
<span class="c1"># self-attention</span>
<span class="n">residual</span> <span class="o">=</span> <span class="n">hidden_states</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">residual_scaling</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/1.0: hidden_states&#39;</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">input_layernorm</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/2.1: normed attn_input&#39;</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="n">attention_output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">self_attention</span><span class="p">(</span>
<span class="n">hidden_states</span><span class="o">=</span><span class="n">hidden_states</span><span class="p">,</span>
<span class="n">attention_mask</span><span class="o">=</span><span class="n">attention_mask_params</span><span class="o">.</span><span class="n">self_attention_mask</span><span class="p">,</span>
<span class="n">use_cache</span><span class="o">=</span><span class="n">use_cache</span><span class="p">,</span>
<span class="n">kv_cache_params</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="p">,</span>
<span class="n">attention_params</span><span class="o">=</span><span class="n">attention_params</span><span class="p">,</span>
<span class="n">lora_layer_params</span><span class="o">=</span><span class="n">lora_layer_params</span><span class="p">)</span>
<span class="k">if</span> <span class="n">use_cache</span><span class="p">:</span>
<span class="n">attention_output</span><span class="p">,</span> <span class="n">presents_self</span> <span class="o">=</span> <span class="n">attention_output</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">attention_output</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/3.1: self_attention_output&#39;</span><span class="p">,</span>
<span class="n">attention_output</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">residual</span> <span class="o">+</span> <span class="n">attention_output</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/3.1: attention_output_with_residual&#39;</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">fp16_clamping</span><span class="p">:</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">maximum</span><span class="p">(</span><span class="o">-</span><span class="mf">64000.0</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">minimum</span><span class="p">(</span><span class="mf">64000.0</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">)</span>
<span class="c1"># MLP</span>
<span class="n">residual</span> <span class="o">=</span> <span class="n">hidden_states</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">residual_scaling</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">post_layernorm</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/3.2: normed_mlp_input&#39;</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">mlp</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">,</span>
<span class="n">lora_layer_params</span><span class="o">=</span><span class="n">lora_layer_params</span><span class="p">)</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/4.1: mlp_output&#39;</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">residual</span> <span class="o">+</span> <span class="n">hidden_states</span>
<span class="k">if</span> <span class="n">ADD_DEBUG_TENSOR</span><span class="p">:</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="bp">self</span><span class="o">.</span><span class="n">local_layer_idx</span><span class="si">:</span><span class="s1">2d</span><span class="si">}</span><span class="s1">/4.2: mlp_output_residual&#39;</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">fp16_clamping</span><span class="p">:</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">maximum</span><span class="p">(</span><span class="o">-</span><span class="mf">64000.0</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">minimum</span><span class="p">(</span><span class="mf">64000.0</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">)</span>
<span class="k">if</span> <span class="n">use_cache</span><span class="p">:</span>
<span class="k">return</span> <span class="p">(</span><span class="n">hidden_states</span><span class="p">,</span> <span class="n">presents_self</span><span class="p">)</span>
<span class="k">return</span> <span class="n">hidden_states</span>
<div class="viewcode-block" id="MLLaMAModel">
<a class="viewcode-back" href="../../../../python-api/tensorrt_llm.models.html#tensorrt_llm.models.MLLaMAModel">[docs]</a>
<span class="k">class</span> <span class="nc">MLLaMAModel</span><span class="p">(</span><span class="n">PretrainedModel</span><span class="p">):</span>
<span class="n">config_class</span> <span class="o">=</span> <span class="n">MLLaMAConfig</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">config</span><span class="p">:</span> <span class="n">MLLaMAConfig</span><span class="p">):</span>
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">config</span><span class="p">)</span>
<span class="n">Attention</span><span class="o">.</span><span class="n">create_attention_const_params</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">config</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">position_embedding_type</span> <span class="o">=</span> <span class="n">config</span><span class="o">.</span><span class="n">position_embedding_type</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mapping</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">mapping</span>
<span class="n">type_vocab_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">type_vocab_size</span>
<span class="bp">self</span><span class="o">.</span><span class="n">has_token_type_embedding</span> <span class="o">=</span> <span class="p">(</span><span class="n">type_vocab_size</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">rescale_before_lm_head</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">rescale_before_lm_head</span>
<span class="bp">self</span><span class="o">.</span><span class="n">layernorm_type</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">layernorm_type</span>
<span class="n">ln_type</span> <span class="o">=</span> <span class="n">layernorm_map</span><span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">layernorm_type</span><span class="p">]</span>
<span class="bp">self</span><span class="o">.</span><span class="n">has_attention_qkvo_bias</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">has_attention_qkvo_bias</span>
<span class="bp">self</span><span class="o">.</span><span class="n">has_mlp_bias</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">has_mlp_bias</span>
<span class="bp">self</span><span class="o">.</span><span class="n">has_model_final_layernorm</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">has_model_final_layernorm</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_dtype</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">dtype</span>
<span class="c1"># no quantization considered for now</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_kv_dtype</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_dtype</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_logits_dtype</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">logits_dtype</span>
<span class="bp">self</span><span class="o">.</span><span class="n">total_num_layers</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">num_hidden_layers</span>
<span class="bp">self</span><span class="o">.</span><span class="n">num_layers</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">num_hidden_layers</span> <span class="o">//</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">pp_size</span>
<span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">hidden_size</span>
<span class="bp">self</span><span class="o">.</span><span class="n">encoder_hidden_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">hidden_size</span>
<span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">num_attention_heads</span>
<span class="c1"># num_kv_heads = self.num_heads</span>
<span class="n">num_kv_heads</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">num_key_value_heads</span>
<span class="k">if</span> <span class="n">num_kv_heads</span> <span class="ow">is</span> <span class="kc">None</span> <span class="ow">or</span> <span class="n">num_kv_heads</span> <span class="o">&lt;=</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">num_kv_heads</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span>
<span class="bp">self</span><span class="o">.</span><span class="n">num_kv_heads</span> <span class="o">=</span> <span class="n">num_kv_heads</span>
<span class="bp">self</span><span class="o">.</span><span class="n">head_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span> <span class="o">//</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">head_size</span> <span class="ow">is</span> <span class="kc">None</span> <span class="k">else</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">head_size</span>
<span class="bp">self</span><span class="o">.</span><span class="n">has_token_type_embedding</span> <span class="o">=</span> <span class="n">type_vocab_size</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">fp16_clamping</span> <span class="o">=</span> <span class="kc">False</span>
<span class="bp">self</span><span class="o">.</span><span class="n">skip_cross_kv</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">skip_cross_kv</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mlp_type</span> <span class="o">=</span> <span class="n">MLPType</span><span class="o">.</span><span class="n">MLP</span> <span class="k">if</span> <span class="ow">not</span> <span class="nb">hasattr</span><span class="p">(</span>
<span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="p">,</span> <span class="s2">&quot;mlp_type&quot;</span><span class="p">)</span> <span class="k">else</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">mlp_type</span>
<span class="bp">self</span><span class="o">.</span><span class="n">use_implicit_relative_attention</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">use_implicit_relative_attention</span> <span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span>
<span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="p">,</span> <span class="s2">&quot;use_implicit_relative_attention&quot;</span><span class="p">)</span> <span class="k">else</span> <span class="kc">False</span>
<span class="bp">self</span><span class="o">.</span><span class="n">cross_attention_layers</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">cross_attention_layers</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_first_pp_rank</span><span class="p">():</span>
<span class="bp">self</span><span class="o">.</span><span class="n">vocab_embedding</span> <span class="o">=</span> <span class="n">Embedding</span><span class="p">(</span>
<span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">embed_vocab_size</span><span class="p">,</span>
<span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_dtype</span><span class="p">,</span>
<span class="n">tp_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">use_parallel_embedding</span> <span class="k">else</span> <span class="mi">1</span><span class="p">,</span>
<span class="n">tp_group</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_group</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">use_parallel_embedding</span> <span class="k">else</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">sharding_dim</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">embedding_sharding_dim</span><span class="p">,</span>
<span class="n">tp_rank</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_rank</span><span class="p">)</span>
<span class="n">layers_range</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">pp_layers</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">total_num_layers</span><span class="p">)</span>
<span class="n">_layers</span> <span class="o">=</span> <span class="p">[]</span>
<span class="k">for</span> <span class="n">layer_idx</span> <span class="ow">in</span> <span class="n">layers_range</span><span class="p">:</span>
<span class="n">local_layer_idx</span> <span class="o">=</span> <span class="n">layer_idx</span> <span class="o">-</span> <span class="n">layers_range</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="n">args</span> <span class="o">=</span> <span class="p">{</span>
<span class="s2">&quot;local_layer_idx&quot;</span><span class="p">:</span> <span class="n">local_layer_idx</span><span class="p">,</span>
<span class="s2">&quot;hidden_size&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
<span class="s2">&quot;ffn_hidden_size&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">intermediate_size</span><span class="p">,</span>
<span class="s2">&quot;num_attention_heads&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span><span class="p">,</span>
<span class="s2">&quot;num_kv_heads&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">num_kv_heads</span><span class="p">,</span>
<span class="s2">&quot;head_size&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">head_size</span><span class="p">,</span>
<span class="s2">&quot;max_position_embeddings&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">max_position_embeddings</span><span class="p">,</span>
<span class="s2">&quot;layernorm_position&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">layernorm_position</span><span class="p">,</span>
<span class="s2">&quot;layernorm_eps&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">norm_epsilon</span><span class="p">,</span>
<span class="s2">&quot;layernorm_type&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">layernorm_type</span><span class="p">,</span>
<span class="s2">&quot;hidden_act&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">hidden_act</span><span class="p">,</span>
<span class="s2">&quot;mlp_type&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">mlp_type</span><span class="p">,</span>
<span class="s2">&quot;mapping&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="p">,</span>
<span class="s2">&quot;dtype&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">_dtype</span><span class="p">,</span>
<span class="s2">&quot;residual_scaling&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">residual_scaling</span><span class="p">,</span>
<span class="s2">&quot;max_distance&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">max_distance</span><span class="p">,</span>
<span class="s2">&quot;num_buckets&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">num_buckets</span><span class="p">,</span>
<span class="s2">&quot;fp16_clamping&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">fp16_clamping</span><span class="p">,</span>
<span class="s2">&quot;skip_cross_kv&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">skip_cross_kv</span><span class="p">,</span>
<span class="s2">&quot;rotary_embedding_base&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">rotary_base</span><span class="p">,</span>
<span class="s2">&quot;rotary_embedding_scaling&quot;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">rotary_scaling</span><span class="p">,</span>
<span class="p">}</span>
<span class="k">if</span> <span class="n">layer_idx</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">cross_attention_layers</span><span class="p">:</span>
<span class="k">assert</span> <span class="n">layers_range</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">==</span> <span class="mi">0</span><span class="p">,</span> <span class="s2">&quot;not support PP now&quot;</span>
<span class="n">_layers</span><span class="o">.</span><span class="n">append</span><span class="p">(</span>
<span class="n">CrossAttentionTransformerBlock</span><span class="p">(</span>
<span class="o">**</span><span class="n">args</span><span class="p">,</span>
<span class="n">layer_idx_in_cache_pool</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span>
<span class="n">num_kv_heads_per_cross_attn_layer</span><span class="p">[:</span><span class="n">local_layer_idx</span><span class="p">]</span><span class="o">.</span>
<span class="n">count</span><span class="p">(</span><span class="n">num_kv_heads</span><span class="p">)))</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">_layers</span><span class="o">.</span><span class="n">append</span><span class="p">(</span>
<span class="n">TransformerBlock</span><span class="p">(</span><span class="o">**</span><span class="n">args</span><span class="p">,</span>
<span class="n">layer_idx_in_cache_pool</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span>
<span class="n">num_kv_heads_per_layer</span><span class="p">[:</span><span class="n">local_layer_idx</span><span class="p">]</span><span class="o">.</span>
<span class="n">count</span><span class="p">(</span><span class="n">num_kv_heads</span><span class="p">)))</span>
<span class="bp">self</span><span class="o">.</span><span class="n">decoder_layers</span> <span class="o">=</span> <span class="n">ModuleList</span><span class="p">(</span><span class="n">_layers</span><span class="p">)</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_last_pp_rank</span><span class="p">():</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">has_model_final_layernorm</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">ln_f</span> <span class="o">=</span> <span class="n">ln_type</span><span class="p">(</span><span class="n">normalized_shape</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
<span class="n">eps</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">norm_epsilon</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">lm_head</span> <span class="o">=</span> <span class="n">ColumnLinear</span><span class="p">(</span>
<span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">,</span>
<span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">vocab_size</span><span class="p">,</span>
<span class="n">bias</span><span class="o">=</span><span class="kc">False</span> <span class="k">if</span> <span class="ow">not</span> <span class="nb">hasattr</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="p">,</span> <span class="s2">&quot;has_lm_head_bias&quot;</span><span class="p">)</span> <span class="k">else</span>
<span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">has_lm_head_bias</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
<span class="n">tp_group</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_group</span><span class="p">,</span>
<span class="n">tp_size</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="n">gather_output</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">trtllm_modules_to_hf_modules</span> <span class="o">=</span> <span class="p">{</span>
<span class="o">**</span><span class="n">get_default_trtllm_modules_to_hf_modules</span><span class="p">(),</span>
<span class="s2">&quot;attn_q&quot;</span><span class="p">:</span> <span class="s2">&quot;self_attn.q_proj&quot;</span><span class="p">,</span>
<span class="s2">&quot;attn_k&quot;</span><span class="p">:</span> <span class="s2">&quot;self_attn.k_proj&quot;</span><span class="p">,</span>
<span class="s2">&quot;attn_v&quot;</span><span class="p">:</span> <span class="s2">&quot;self_attn.v_proj&quot;</span><span class="p">,</span>
<span class="s2">&quot;attn_dense&quot;</span><span class="p">:</span> <span class="s2">&quot;self_attn.o_proj&quot;</span><span class="p">,</span>
<span class="s2">&quot;cross_attn_q&quot;</span><span class="p">:</span> <span class="s2">&quot;encoder_attn.q_proj&quot;</span><span class="p">,</span>
<span class="s2">&quot;cross_attn_k&quot;</span><span class="p">:</span> <span class="s2">&quot;encoder_attn.k_proj&quot;</span><span class="p">,</span>
<span class="s2">&quot;cross_attn_v&quot;</span><span class="p">:</span> <span class="s2">&quot;encoder_attn.v_proj&quot;</span><span class="p">,</span>
<span class="s2">&quot;cross_attn_dense&quot;</span><span class="p">:</span> <span class="s2">&quot;encoder_attn.o_proj&quot;</span><span class="p">,</span>
<span class="p">}</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">relative_attention</span> <span class="ow">and</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">use_implicit_relative_attention</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">rel_attn_table</span> <span class="o">=</span> <span class="n">Parameter</span><span class="p">(</span>
<span class="n">shape</span><span class="o">=</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">num_attention_heads</span> <span class="o">//</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">num_buckets</span><span class="p">),</span>
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_dtype</span><span class="p">)</span>
<div class="viewcode-block" id="MLLaMAModel.forward">
<a class="viewcode-back" href="../../../../python-api/tensorrt_llm.models.html#tensorrt_llm.models.MLLaMAModel.forward">[docs]</a>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span>
<span class="bp">self</span><span class="p">,</span>
<span class="n">decoder_input_ids</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
<span class="n">encoder_output</span><span class="p">:</span> <span class="n">Tensor</span><span class="p">,</span>
<span class="n">use_cache</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">attention_mask_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">last_token_ids</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">kv_cache_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">attention_params</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="n">lora_params</span><span class="p">:</span> <span class="n">LoraParams</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">cross_kv_cache_gen</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">cross_kv_reuse</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">prompt_embedding_table</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">prompt_tasks</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">prompt_vocab_size</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">skip_cross_attn_blocks</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Tensor</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="p">):</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_first_pp_rank</span><span class="p">():</span>
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">decoder_input_ids</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">,</span> <span class="n">Tensor</span><span class="p">)</span>
<span class="n">attention_params</span> <span class="o">=</span> <span class="n">Attention</span><span class="o">.</span><span class="n">fill_attention_params</span><span class="p">(</span>
<span class="bp">self</span><span class="p">,</span> <span class="n">attention_params</span><span class="p">)</span>
<span class="c1"># In PP, layer 0 has ids as inputs, all other layers have hidden_states as inputs</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_first_pp_rank</span><span class="p">():</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">vocab_embedding</span><span class="p">(</span><span class="n">decoder_input_ids</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">register_network_output</span><span class="p">(</span><span class="s1">&#39;embedding_layer_output&#39;</span><span class="p">,</span>
<span class="n">hidden_states</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">recv</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">prev_pp_rank</span><span class="p">())</span>
<span class="n">kv_cache_params</span><span class="o">.</span><span class="n">fill_none_tensor_list</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">decoder_layers</span><span class="p">))</span>
<span class="n">full_text_row_masked_out_mask</span> <span class="o">=</span> <span class="n">reduce</span><span class="p">(</span>
<span class="p">(</span><span class="n">attention_mask_params</span><span class="o">.</span><span class="n">cross_attention_mask</span><span class="p">)</span><span class="o">.</span><span class="n">cast</span><span class="p">(</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">dtype</span><span class="p">),</span>
<span class="n">trt</span><span class="o">.</span><span class="n">ReduceOperation</span><span class="o">.</span><span class="n">MAX</span><span class="p">,</span>
<span class="n">dim</span><span class="o">=-</span><span class="mi">1</span><span class="p">,</span>
<span class="n">keepdim</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="n">cross_attention_mask_type</span> <span class="o">=</span> <span class="n">attention_mask_params</span><span class="o">.</span><span class="n">cross_attention_mask</span><span class="o">.</span><span class="n">dtype</span>
<span class="n">attention_mask_params</span><span class="o">.</span><span class="n">cross_attention_mask</span> <span class="o">=</span> <span class="p">(</span>
<span class="n">attention_mask_params</span><span class="o">.</span><span class="n">cross_attention_mask</span><span class="o">.</span><span class="n">cast</span><span class="p">(</span>
<span class="n">full_text_row_masked_out_mask</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span> <span class="o">*</span>
<span class="n">full_text_row_masked_out_mask</span><span class="p">)</span><span class="o">.</span><span class="n">cast</span><span class="p">(</span><span class="n">cross_attention_mask_type</span><span class="p">)</span>
<span class="k">if</span> <span class="n">use_cache</span><span class="p">:</span>
<span class="n">presents</span> <span class="o">=</span> <span class="p">[]</span>
<span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="p">(</span><span class="n">decoder_layer</span><span class="p">,</span> <span class="n">past</span><span class="p">)</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span>
<span class="nb">zip</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">decoder_layers</span><span class="p">,</span> <span class="n">kv_cache_params</span><span class="o">.</span><span class="n">past_key_value</span><span class="p">)):</span>
<span class="n">lora_layer_params</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">lora_params</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">lora_params</span><span class="o">.</span><span class="n">lora_ranks</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">lora_layer_params</span> <span class="o">=</span> <span class="n">lora_params</span><span class="o">.</span><span class="n">get_layer_params</span><span class="p">(</span><span class="n">i</span><span class="p">)</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">decoder_layer</span><span class="p">(</span>
<span class="n">hidden_states</span><span class="p">,</span>
<span class="n">encoder_output</span><span class="o">=</span><span class="n">encoder_output</span><span class="p">,</span>
<span class="n">attention_mask_params</span><span class="o">=</span><span class="n">attention_mask_params</span><span class="p">,</span>
<span class="n">use_cache</span><span class="o">=</span><span class="n">use_cache</span><span class="p">,</span>
<span class="n">kv_cache_params</span><span class="o">=</span><span class="n">KeyValueCacheParams</span><span class="p">(</span>
<span class="n">past_key_value</span><span class="o">=</span><span class="n">past</span><span class="p">,</span>
<span class="n">host_past_key_value_lengths</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="o">.</span>
<span class="n">host_past_key_value_lengths</span><span class="p">,</span>
<span class="n">host_max_attention_window_sizes</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="o">.</span>
<span class="n">host_max_attention_window_sizes</span><span class="p">,</span>
<span class="n">host_sink_token_length</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="o">.</span>
<span class="n">host_sink_token_length</span><span class="p">,</span>
<span class="n">cache_indirection</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="o">.</span><span class="n">cache_indirection</span><span class="p">,</span>
<span class="n">kv_cache_block_offsets</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="o">.</span>
<span class="n">kv_cache_block_offsets</span><span class="p">,</span>
<span class="n">host_kv_cache_block_offsets</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="o">.</span>
<span class="n">host_cross_kv_cache_block_offsets</span><span class="p">,</span>
<span class="n">host_kv_cache_pool_pointers</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="o">.</span>
<span class="n">host_kv_cache_pool_pointers</span><span class="p">,</span>
<span class="n">host_kv_cache_pool_mapping</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="o">.</span>
<span class="n">host_kv_cache_pool_mapping</span><span class="p">,</span>
<span class="n">cross_kv_cache_block_offsets</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="o">.</span>
<span class="n">cross_kv_cache_block_offsets</span><span class="p">,</span>
<span class="n">host_cross_kv_cache_block_offsets</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="o">.</span>
<span class="n">host_cross_kv_cache_block_offsets</span><span class="p">,</span>
<span class="n">host_cross_kv_cache_pool_pointers</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="o">.</span>
<span class="n">host_cross_kv_cache_pool_pointers</span><span class="p">,</span>
<span class="n">host_cross_kv_cache_pool_mapping</span><span class="o">=</span><span class="n">kv_cache_params</span><span class="o">.</span>
<span class="n">host_cross_kv_cache_pool_mapping</span><span class="p">,</span>
<span class="p">),</span>
<span class="n">skip_cross_attn_blocks</span><span class="o">=</span><span class="n">skip_cross_attn_blocks</span> <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span>
<span class="n">decoder_layer</span><span class="p">,</span> <span class="n">CrossAttentionTransformerBlock</span><span class="p">)</span> <span class="k">else</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">attention_params</span><span class="o">=</span><span class="n">attention_params</span><span class="p">,</span>
<span class="n">lora_layer_params</span><span class="o">=</span><span class="n">lora_layer_params</span><span class="p">,</span>
<span class="n">cross_kv_cache_gen</span><span class="o">=</span><span class="n">cross_kv_cache_gen</span><span class="p">,</span>
<span class="n">cross_kv_reuse</span><span class="o">=</span><span class="n">cross_kv_reuse</span><span class="p">,</span>
<span class="n">full_text_row_masked_out_mask</span><span class="o">=</span><span class="n">full_text_row_masked_out_mask</span><span class="p">,</span>
<span class="p">)</span>
<span class="k">if</span> <span class="n">use_cache</span><span class="p">:</span>
<span class="n">present</span> <span class="o">=</span> <span class="n">hidden_states</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span>
<span class="n">presents</span><span class="o">.</span><span class="n">append</span><span class="p">((</span><span class="n">present</span><span class="p">))</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">hidden_states</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_last_pp_rank</span><span class="p">():</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">has_model_final_layernorm</span><span class="p">:</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">ln_f</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="c1"># [bs, seq, hidden_size] or [num_tokens, hidden_size] -&gt; [bs, hidden_size]</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">gather_last_token_logits</span><span class="p">(</span>
<span class="n">hidden_states</span><span class="p">,</span> <span class="n">last_token_ids</span><span class="p">,</span>
<span class="n">default_net</span><span class="p">()</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">remove_input_padding</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">register_network_output</span><span class="p">(</span><span class="s1">&#39;logits_before_lmhead&#39;</span><span class="p">,</span> <span class="n">hidden_states</span><span class="p">)</span>
<span class="c1"># [bs, hidden_size] -&gt; [bs, vocab_size]</span>
<span class="n">lm_logits</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">lm_head</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">)</span>
<span class="n">lm_logits</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;logits&#39;</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">_logits_dtype</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">send</span><span class="p">(</span><span class="n">hidden_states</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">next_pp_rank</span><span class="p">())</span>
<span class="n">hidden_states</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;hidden_states_output&#39;</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">_dtype</span><span class="p">)</span>
<span class="k">if</span> <span class="n">use_cache</span> <span class="ow">and</span> <span class="n">default_net</span><span class="p">()</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">paged_kv_cache</span> <span class="o">==</span> <span class="kc">False</span><span class="p">:</span>
<span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">present</span> <span class="ow">in</span> <span class="nb">zip</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">pp_layers</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">total_num_layers</span><span class="p">),</span>
<span class="n">presents</span><span class="p">):</span>
<span class="n">present</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;present_key_value_</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">_kv_dtype</span><span class="p">)</span>
<span class="k">if</span> <span class="n">default_net</span><span class="p">()</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">gpt_attention_plugin</span><span class="p">:</span>
<span class="n">present</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span><span class="o">.</span><span class="n">mark_output</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;cross_present_key_value_</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">,</span>
<span class="bp">self</span><span class="o">.</span><span class="n">_kv_dtype</span><span class="p">)</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_last_pp_rank</span><span class="p">():</span>
<span class="k">return</span> <span class="p">(</span><span class="n">lm_logits</span><span class="p">,</span> <span class="nb">tuple</span><span class="p">(</span><span class="n">presents</span><span class="p">))</span>
<span class="k">return</span> <span class="p">(</span><span class="n">hidden_states</span><span class="p">,</span> <span class="nb">tuple</span><span class="p">(</span><span class="n">presents</span><span class="p">))</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_last_pp_rank</span><span class="p">():</span>
<span class="k">return</span> <span class="n">lm_logits</span>
<span class="k">return</span> <span class="n">hidden_states</span></div>
<div class="viewcode-block" id="MLLaMAModel.prepare_inputs">
<a class="viewcode-back" href="../../../../python-api/tensorrt_llm.models.html#tensorrt_llm.models.MLLaMAModel.prepare_inputs">[docs]</a>
<span class="k">def</span> <span class="nf">prepare_inputs</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span>
<span class="n">max_batch_size</span><span class="p">,</span>
<span class="n">max_beam_width</span><span class="p">,</span>
<span class="n">max_decoder_input_len</span><span class="p">,</span>
<span class="n">max_seq_len</span><span class="p">,</span>
<span class="n">max_encoder_input_len</span><span class="p">,</span>
<span class="n">gather_context_logits</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
<span class="n">gather_generation_logits</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">,</span>
<span class="n">lora_target_modules</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">prompt_embedding_table_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">0</span><span class="p">,</span>
<span class="n">use_cache</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="o">*</span><span class="n">args</span><span class="p">,</span>
<span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
<span class="w"> </span><span class="sd">&#39;&#39;&#39;@brief: Prepare inputs Tensors for the model, the given sizes are used to determine the</span>
<span class="sd"> ranges of the dimensions of when using TRT dynamic shapes.</span>
<span class="sd"> @return: a list contains values which can be fed into the self.forward()</span>
<span class="sd"> &#39;&#39;&#39;</span>
<span class="c1"># Prepare inputs</span>
<span class="n">max_output_len</span> <span class="o">=</span> <span class="n">max_decoder_input_len</span> <span class="o">+</span> <span class="n">max_seq_len</span>
<span class="n">head_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">head_size</span>
<span class="n">num_kv_heads</span> <span class="o">=</span> <span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">num_kv_heads</span> <span class="o">+</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span> <span class="o">-</span>
<span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span>
<span class="c1"># TODO check</span>
<span class="c1"># encoder_head_size = self.encoder_head_size</span>
<span class="c1"># encoder_num_kv_heads = (self.encoder_num_kv_heads + self.mapping.tp_size</span>
<span class="c1"># - 1) // self.mapping.tp_size</span>
<span class="n">encoder_head_size</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">head_size</span>
<span class="n">encoder_num_kv_heads</span> <span class="o">=</span> <span class="n">num_kv_heads</span>
<span class="n">bb_range</span> <span class="o">=</span> <span class="p">[</span>
<span class="mi">1</span><span class="p">,</span> <span class="p">(</span><span class="n">max_batch_size</span> <span class="o">*</span> <span class="n">max_beam_width</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="mi">2</span><span class="p">,</span>
<span class="n">max_batch_size</span> <span class="o">*</span> <span class="n">max_beam_width</span>
<span class="p">]</span>
<span class="n">bs_range</span> <span class="o">=</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="p">(</span><span class="n">max_batch_size</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="mi">2</span><span class="p">,</span> <span class="n">max_batch_size</span><span class="p">]</span>
<span class="n">beam_width_range</span> <span class="o">=</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="p">(</span><span class="n">max_beam_width</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="mi">2</span><span class="p">,</span> <span class="n">max_beam_width</span><span class="p">]</span>
<span class="n">inlen_range</span> <span class="o">=</span> <span class="p">[</span>
<span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="n">max_decoder_input_len</span>
<span class="p">]</span> <span class="c1"># context phase &gt;= 1 (if forced_input_ids), generation phase = 1</span>
<span class="n">encoder_inlen_range</span> <span class="o">=</span> <span class="p">[</span>
<span class="mi">1</span><span class="p">,</span> <span class="p">(</span><span class="n">max_encoder_input_len</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="mi">2</span><span class="p">,</span> <span class="n">max_encoder_input_len</span>
<span class="p">]</span>
<span class="n">mask_len_range</span> <span class="o">=</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="p">(</span><span class="n">max_output_len</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="mi">2</span> <span class="o">+</span> <span class="mi">1</span><span class="p">,</span> <span class="n">max_output_len</span> <span class="o">+</span> <span class="mi">1</span><span class="p">]</span>
<span class="n">max_output_len_range</span> <span class="o">=</span> <span class="p">[</span><span class="mi">0</span><span class="p">,</span> <span class="p">(</span><span class="n">max_output_len</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="mi">2</span><span class="p">,</span> <span class="n">max_output_len</span><span class="p">]</span>
<span class="n">encoder_num_tokens_range</span> <span class="o">=</span> <span class="p">[</span>
<span class="mi">0</span><span class="p">,</span> <span class="c1"># 0 for generation phase, &gt;0 for context phase</span>
<span class="p">(</span><span class="n">max_encoder_input_len</span> <span class="o">*</span> <span class="n">max_batch_size</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="mi">2</span><span class="p">,</span>
<span class="n">max_encoder_input_len</span> <span class="o">*</span> <span class="n">max_batch_size</span><span class="p">,</span>
<span class="p">]</span>
<span class="n">decoder_num_tokens_range</span> <span class="o">=</span> <span class="p">[</span>
<span class="mi">1</span><span class="p">,</span>
<span class="n">max_batch_size</span> <span class="o">*</span> <span class="n">max_beam_width</span><span class="p">,</span>
<span class="nb">max</span><span class="p">(</span><span class="n">max_decoder_input_len</span> <span class="o">*</span> <span class="n">max_batch_size</span><span class="p">,</span>
<span class="n">max_beam_width</span> <span class="o">*</span> <span class="n">max_batch_size</span><span class="p">),</span>
<span class="p">]</span>
<span class="c1"># No enable_two_optimization_profiles support yet</span>
<span class="n">encoder_input_len_range</span> <span class="o">=</span> <span class="p">[</span>
<span class="mi">0</span><span class="p">,</span> <span class="c1"># 0 for generation phase, &gt;0 for context phase</span>
<span class="p">(</span><span class="n">max_encoder_input_len</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="mi">2</span><span class="p">,</span>
<span class="n">max_encoder_input_len</span>
<span class="p">]</span>
<span class="c1"># pack masks into bits (store as int32).</span>
<span class="n">max_cross_packed_mask_dim0</span> <span class="o">=</span> <span class="n">max_batch_size</span> <span class="o">*</span> <span class="p">(</span>
<span class="p">(</span><span class="n">max_decoder_input_len</span> <span class="o">+</span> <span class="mi">128</span> <span class="o">-</span> <span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="mi">128</span><span class="p">)</span> <span class="o">*</span> <span class="mi">128</span>
<span class="n">max_cross_packed_mask_dim1</span> <span class="o">=</span> <span class="p">(</span>
<span class="p">(</span><span class="n">max_encoder_input_len</span> <span class="o">+</span> <span class="mi">256</span> <span class="o">-</span> <span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="mi">256</span><span class="p">)</span> <span class="o">*</span> <span class="mi">256</span> <span class="o">//</span> <span class="mi">32</span>
<span class="n">cross_packed_mask_dim0_range</span> <span class="o">=</span> <span class="p">[</span>
<span class="mi">1</span><span class="p">,</span> <span class="p">(</span><span class="n">max_cross_packed_mask_dim0</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="mi">2</span><span class="p">,</span> <span class="n">max_cross_packed_mask_dim0</span>
<span class="p">]</span>
<span class="n">cross_packed_mask_dim1_range</span> <span class="o">=</span> <span class="p">[</span>
<span class="mi">0</span><span class="p">,</span> <span class="c1"># 0 for generation phase, &gt;0 for context phase</span>
<span class="p">(</span><span class="n">max_cross_packed_mask_dim1</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="mi">2</span><span class="p">,</span>
<span class="n">max_cross_packed_mask_dim1</span>
<span class="p">]</span>
<span class="n">past_key_value</span> <span class="o">=</span> <span class="p">[]</span>
<span class="n">sequence_length</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">host_past_key_value_lengths</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">attention_mask_params</span> <span class="o">=</span> <span class="n">AttentionMaskParams</span><span class="p">()</span>
<span class="n">use_gpt_attention_plugin</span> <span class="o">=</span> <span class="n">default_net</span><span class="p">(</span>
<span class="p">)</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">gpt_attention_plugin</span>
<span class="n">remove_input_padding</span> <span class="o">=</span> <span class="n">default_net</span><span class="p">()</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">remove_input_padding</span>
<span class="n">paged_kv_cache</span> <span class="o">=</span> <span class="n">default_net</span><span class="p">()</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">paged_kv_cache</span>
<span class="n">tokens_per_block</span> <span class="o">=</span> <span class="n">default_net</span><span class="p">()</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">tokens_per_block</span>
<span class="n">use_lora_plugin</span> <span class="o">=</span> <span class="n">default_net</span><span class="p">()</span><span class="o">.</span><span class="n">plugin_config</span><span class="o">.</span><span class="n">lora_plugin</span>
<span class="n">kv_cache_type</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">use_cache</span><span class="p">:</span>
<span class="n">kv_cache_type</span> <span class="o">=</span> <span class="n">KVCacheType</span><span class="o">.</span><span class="n">DISABLED</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">if</span> <span class="n">paged_kv_cache</span><span class="p">:</span>
<span class="n">kv_cache_type</span> <span class="o">=</span> <span class="n">KVCacheType</span><span class="o">.</span><span class="n">PAGED</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">kv_cache_type</span> <span class="o">=</span> <span class="n">KVCacheType</span><span class="o">.</span><span class="n">CONTINUOUS</span>
<span class="n">input_ids</span><span class="p">,</span> <span class="n">hidden_states</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">remove_input_padding</span><span class="p">:</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_first_pp_rank</span><span class="p">():</span>
<span class="n">input_ids</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;input_ids&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;decoder_num_tokens&#39;</span><span class="p">,</span>
<span class="p">[</span><span class="n">decoder_num_tokens_range</span><span class="p">]),</span>
<span class="p">]))</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;hidden_states_input&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_dtype</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;decoder_num_tokens&#39;</span><span class="p">,</span>
<span class="p">[</span><span class="n">decoder_num_tokens_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;hidden_size&#39;</span><span class="p">,</span> <span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">]),</span>
<span class="p">]))</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_first_pp_rank</span><span class="p">():</span>
<span class="n">input_ids</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;input_ids&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;input_len&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">inlen_range</span><span class="p">]),</span>
<span class="p">]))</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">hidden_states</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;hidden_states_input&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_dtype</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span>
<span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;input_len&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">inlen_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;hidden_size&#39;</span><span class="p">,</span> <span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">]),</span>
<span class="p">]))</span>
<span class="n">encoder_input_lengths</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s2">&quot;encoder_input_lengths&quot;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([(</span><span class="s2">&quot;batch_size_beam_width&quot;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">])]),</span>
<span class="p">)</span>
<span class="n">encoder_max_input_length</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s2">&quot;encoder_max_input_length&quot;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([(</span><span class="s2">&quot;encoder_max_input_length&quot;</span><span class="p">,</span>
<span class="p">[</span><span class="n">encoder_inlen_range</span><span class="p">])]),</span>
<span class="p">)</span>
<span class="k">if</span> <span class="n">remove_input_padding</span><span class="p">:</span>
<span class="n">encoder_output</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s2">&quot;encoder_output&quot;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_dtype</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s2">&quot;encoder_num_tokens&quot;</span><span class="p">,</span> <span class="p">[</span><span class="n">encoder_num_tokens_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s2">&quot;hidden_size&quot;</span><span class="p">,</span> <span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">]),</span>
<span class="p">]),</span>
<span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">encoder_output</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s2">&quot;encoder_output&quot;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_dtype</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s2">&quot;batch_size_beam_width_encoder&quot;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s2">&quot;encoder_input_len&quot;</span><span class="p">,</span> <span class="p">[</span><span class="n">encoder_input_len_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s2">&quot;hidden_size&quot;</span><span class="p">,</span> <span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">]),</span>
<span class="p">]),</span>
<span class="p">)</span>
<span class="n">context_lengths</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">host_context_lengths</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">host_request_types</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">host_runtime_perf_knobs</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">host_context_progress</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">use_gpt_attention_plugin</span> <span class="ow">and</span> <span class="n">remove_input_padding</span><span class="p">:</span>
<span class="n">host_context_lengths</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;host_context_lengths&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span>
<span class="p">[</span><span class="n">bb_range</span><span class="p">])</span>
<span class="p">]))</span>
<span class="k">if</span> <span class="n">use_gpt_attention_plugin</span><span class="p">:</span>
<span class="k">if</span> <span class="n">kv_cache_type</span> <span class="o">!=</span> <span class="n">KVCacheType</span><span class="o">.</span><span class="n">DISABLED</span><span class="p">:</span>
<span class="n">sequence_length</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s1">&#39;sequence_length&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">])</span>
<span class="p">]),</span>
<span class="p">)</span>
<span class="n">host_past_key_value_lengths</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s1">&#39;host_past_key_value_lengths&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">])</span>
<span class="p">]),</span>
<span class="p">)</span>
<span class="n">context_lengths</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;context_lengths&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">])</span>
<span class="p">]))</span>
<span class="n">host_request_types</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;host_request_types&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span>
<span class="p">[</span><span class="n">bb_range</span><span class="p">])</span>
<span class="p">]))</span>
<span class="n">host_runtime_perf_knobs</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;host_runtime_perf_knobs&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int64</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">16</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;perf_knob_size&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">16</span><span class="p">])</span>
<span class="p">]))</span>
<span class="n">host_context_progress</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;host_context_progress&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int64</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;context_progress_size&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">1</span><span class="p">])</span>
<span class="p">]))</span>
<span class="n">last_token_ids</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_last_pp_rank</span><span class="p">()</span> <span class="ow">and</span> <span class="ow">not</span> <span class="n">gather_context_logits</span><span class="p">:</span>
<span class="n">last_token_ids</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s2">&quot;last_token_ids&quot;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([(</span><span class="s2">&quot;batch_size_last_token_ids&quot;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">])</span>
<span class="p">]),</span>
<span class="p">)</span>
<span class="n">attention_mask</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">use_gpt_attention_plugin</span><span class="p">:</span>
<span class="n">attention_mask</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s1">&#39;attention_mask&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;mask_len&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">mask_len_range</span><span class="p">]),</span>
<span class="p">]),</span>
<span class="p">)</span>
<span class="k">assert</span> <span class="kc">False</span><span class="p">,</span> <span class="s2">&quot;not support non-attention-plugin case now&quot;</span>
<span class="n">cross_attention_mask</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s1">&#39;cross_attention_mask&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">bool</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;decoder_num_tokens_2&#39;</span><span class="p">,</span>
<span class="p">[</span><span class="n">decoder_num_tokens_range</span>
<span class="p">]),</span> <span class="c1"># TODO (bhsueh) should use same name as input_ids</span>
<span class="p">(</span><span class="s1">&#39;encoder_input_len_2&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">encoder_input_len_range</span><span class="p">]),</span>
<span class="p">]),</span>
<span class="p">)</span>
<span class="n">cross_attention_packed_mask</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s1">&#39;cross_attention_packed_mask&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;cross_packed_mask_dim0&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">cross_packed_mask_dim0_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;cross_packed_mask_dim1&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">cross_packed_mask_dim1_range</span><span class="p">]),</span>
<span class="p">]),</span>
<span class="p">)</span>
<span class="c1"># create the attention_mask_params.</span>
<span class="n">attention_mask_params</span> <span class="o">=</span> <span class="n">AttentionMaskParams</span><span class="p">(</span>
<span class="n">attention_mask</span><span class="p">,</span> <span class="kc">None</span><span class="p">,</span> <span class="n">cross_attention_mask</span><span class="p">,</span>
<span class="n">cross_attention_packed_mask</span><span class="p">)</span>
<span class="n">cache_indirection</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s1">&#39;cache_indirection&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;batch_size_cache&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bs_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">beam_width_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;max_seq_len&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">max_output_len_range</span><span class="p">]),</span>
<span class="p">]),</span>
<span class="p">)</span>
<span class="n">layers_range</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">pp_layers</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">total_num_layers</span><span class="p">)</span>
<span class="n">num_pp_layers</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">layers_range</span><span class="p">)</span>
<span class="n">host_max_attention_window_sizes</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">host_sink_token_length</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">use_gpt_attention_plugin</span><span class="p">:</span>
<span class="n">host_max_attention_window_sizes</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="sa">f</span><span class="s1">&#39;host_max_attention_window_sizes&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="n">num_pp_layers</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([(</span><span class="s1">&#39;num_layers&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">num_pp_layers</span><span class="p">])]))</span>
<span class="n">host_sink_token_length</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;host_sink_token_length&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([(</span><span class="s1">&#39;scalar&#39;</span><span class="p">,</span>
<span class="p">[</span><span class="mi">1</span><span class="p">])]))</span>
<span class="c1"># TODO LoRA for mllama is not verified.</span>
<span class="n">lora_weights_pointers</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">lora_ranks</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">lora_params</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">use_lora_plugin</span><span class="p">:</span>
<span class="n">lora_weights_pointers</span> <span class="o">=</span> <span class="p">[]</span>
<span class="n">lora_ranks</span> <span class="o">=</span> <span class="p">[]</span>
<span class="n">missing_qkv_modules</span> <span class="o">=</span> <span class="p">[]</span>
<span class="k">if</span> <span class="nb">any</span><span class="p">(</span><span class="n">x</span> <span class="ow">in</span> <span class="n">lora_target_modules</span>
<span class="k">for</span> <span class="n">x</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">&quot;attn_q&quot;</span><span class="p">,</span> <span class="s2">&quot;attn_k&quot;</span><span class="p">,</span> <span class="s2">&quot;attn_v&quot;</span><span class="p">]):</span>
<span class="k">for</span> <span class="n">lora_module</span> <span class="ow">in</span> <span class="p">[</span>
<span class="s2">&quot;attn_q&quot;</span><span class="p">,</span>
<span class="s2">&quot;attn_k&quot;</span><span class="p">,</span>
<span class="s2">&quot;attn_v&quot;</span><span class="p">,</span>
<span class="p">]:</span>
<span class="k">if</span> <span class="n">lora_module</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">lora_target_modules</span><span class="p">:</span>
<span class="n">missing_qkv_modules</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">lora_module</span><span class="p">)</span>
<span class="k">if</span> <span class="nb">any</span><span class="p">(</span><span class="n">x</span> <span class="ow">in</span> <span class="n">lora_target_modules</span>
<span class="k">for</span> <span class="n">x</span> <span class="ow">in</span> <span class="p">[</span><span class="s2">&quot;cross_attn_q&quot;</span><span class="p">,</span> <span class="s2">&quot;cross_attn_k&quot;</span><span class="p">,</span> <span class="s2">&quot;cross_attn_v&quot;</span><span class="p">]):</span>
<span class="k">for</span> <span class="n">lora_module</span> <span class="ow">in</span> <span class="p">[</span>
<span class="s2">&quot;cross_attn_q&quot;</span><span class="p">,</span> <span class="s2">&quot;cross_attn_k&quot;</span><span class="p">,</span> <span class="s2">&quot;cross_attn_v&quot;</span>
<span class="p">]:</span>
<span class="k">if</span> <span class="n">lora_module</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">lora_target_modules</span><span class="p">:</span>
<span class="n">missing_qkv_modules</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">lora_module</span><span class="p">)</span>
<span class="c1"># For LoRA</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="n">layers_range</span><span class="p">:</span>
<span class="n">lora_weight_pointer_dict</span> <span class="o">=</span> <span class="p">{}</span>
<span class="n">lora_rank_dict</span> <span class="o">=</span> <span class="p">{}</span>
<span class="k">for</span> <span class="n">lora_module</span> <span class="ow">in</span> <span class="p">(</span><span class="n">lora_target_modules</span> <span class="o">+</span> <span class="n">missing_qkv_modules</span><span class="p">):</span>
<span class="n">lora_weight_pointer</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="n">lora_module</span><span class="si">}</span><span class="s1">_lora_weights_pointers_</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int64</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span>
<span class="p">[</span><span class="n">bb_range</span><span class="p">]),</span> <span class="p">(</span><span class="s1">&#39;in_out&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">2</span><span class="p">])]))</span>
<span class="n">lora_weight_pointer_dict</span><span class="o">.</span><span class="n">update</span><span class="p">({</span>
<span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="n">lora_module</span><span class="si">}</span><span class="s1">_lora_weights_pointers&#39;</span><span class="p">:</span>
<span class="n">lora_weight_pointer</span>
<span class="p">})</span>
<span class="n">lora_rank</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="n">lora_module</span><span class="si">}</span><span class="s1">_lora_ranks_</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">])</span>
<span class="p">]))</span>
<span class="n">lora_rank_dict</span><span class="o">.</span><span class="n">update</span><span class="p">(</span>
<span class="p">{</span><span class="sa">f</span><span class="s1">&#39;</span><span class="si">{</span><span class="n">lora_module</span><span class="si">}</span><span class="s1">_lora_ranks&#39;</span><span class="p">:</span> <span class="n">lora_rank</span><span class="p">})</span>
<span class="n">lora_weights_pointers</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">lora_weight_pointer_dict</span><span class="p">)</span>
<span class="n">lora_ranks</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">lora_rank_dict</span><span class="p">)</span>
<span class="c1"># For cross attention, we need to use encoder_input_lengths (in CPU) to pass</span>
<span class="c1"># as the host_context_lengths to the lora_plugin. But for self attention, we</span>
<span class="c1"># should keep using the original host_context_lengths. Therefore, we keep both</span>
<span class="c1"># of them in the lora_params.</span>
<span class="n">host_encoder_input_lengths</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">remove_input_padding</span><span class="p">:</span>
<span class="n">host_encoder_input_lengths</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s2">&quot;host_encoder_input_lengths&quot;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([(</span><span class="s2">&quot;batch_size_beam_width&quot;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">])</span>
<span class="p">]),</span>
<span class="p">)</span>
<span class="n">lora_params</span> <span class="o">=</span> <span class="n">LoraParams</span><span class="p">(</span>
<span class="n">lora_ranks</span><span class="o">=</span><span class="n">lora_ranks</span><span class="p">,</span>
<span class="n">lora_weights_pointers</span><span class="o">=</span><span class="n">lora_weights_pointers</span><span class="p">,</span>
<span class="n">host_context_lengths</span><span class="o">=</span><span class="n">host_context_lengths</span><span class="p">,</span>
<span class="n">max_context_length</span><span class="o">=</span><span class="n">max_decoder_input_len</span><span class="p">,</span>
<span class="n">max_encoder_context_length</span><span class="o">=</span><span class="n">max_encoder_input_len</span><span class="p">,</span>
<span class="n">host_request_types</span><span class="o">=</span><span class="n">host_request_types</span><span class="p">,</span>
<span class="n">host_encoder_input_lengths</span><span class="o">=</span><span class="n">host_encoder_input_lengths</span><span class="p">,</span>
<span class="p">)</span>
<span class="n">kv_cache_block_offsets</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">host_kv_cache_block_offsets</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">host_kv_cache_pool_pointers</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">host_kv_cache_pool_mapping</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">cross_kv_cache_block_offsets</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">host_cross_kv_cache_block_offsets</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">host_cross_kv_cache_pool_pointers</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">host_cross_kv_cache_pool_mapping</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">use_cache</span><span class="p">:</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">paged_kv_cache</span><span class="p">:</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="n">layers_range</span><span class="p">:</span>
<span class="n">kv_dim_range</span> <span class="o">=</span> <span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;kv&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">2</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;num_heads&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">num_kv_heads</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;past_key_len&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">max_output_len_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;head_size&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">head_size</span><span class="p">]),</span>
<span class="p">])</span>
<span class="n">kv</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="sa">f</span><span class="s1">&#39;past_key_value_</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_kv_dtype</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="n">num_kv_heads</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="n">head_size</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">kv_dim_range</span><span class="p">)</span>
<span class="n">past_key_value</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">kv</span><span class="p">)</span>
<span class="k">if</span> <span class="n">i</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">fusion_schedule</span><span class="p">:</span>
<span class="n">xa_layer_id</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">fusion_schedule</span><span class="o">.</span><span class="n">index</span><span class="p">(</span>
<span class="n">i</span><span class="p">)</span> <span class="o">+</span> <span class="n">layers_range</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">]</span>
<span class="n">cross_kv_dim_range</span> <span class="o">=</span> <span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;kv&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">2</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;cross_num_heads&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">encoder_num_kv_heads</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;cross_past_key_len&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">encoder_input_len_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;cross_head_size&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">encoder_head_size</span><span class="p">]),</span>
<span class="p">])</span>
<span class="n">cross_kv</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="sa">f</span><span class="s1">&#39;cross_past_key_value_</span><span class="si">{</span><span class="n">xa_layer_id</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_kv_dtype</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span>
<span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="n">encoder_num_kv_heads</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="n">encoder_head_size</span>
<span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">cross_kv_dim_range</span><span class="p">)</span>
<span class="n">past_key_value</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">kv</span><span class="p">)</span>
<span class="c1"># TODO: Remove this when TRT fix the named dimension</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">remove_input_padding</span><span class="p">:</span>
<span class="n">assertion</span><span class="p">(</span>
<span class="n">shape</span><span class="p">(</span>
<span class="n">input_ids</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_first_pp_rank</span><span class="p">()</span> <span class="k">else</span>
<span class="n">hidden_states</span><span class="p">,</span> <span class="mi">0</span><span class="p">)</span> <span class="o">==</span> <span class="n">shape</span><span class="p">(</span><span class="n">kv</span><span class="p">,</span> <span class="mi">0</span><span class="p">),</span> <span class="s1">&#39;batch size&#39;</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span> <span class="c1"># paged_kv_cache == True</span>
<span class="c1"># PagedKV setup for KV cache of self-attention</span>
<span class="n">max_blocks_per_seq_range</span> <span class="o">=</span> <span class="p">[[</span>
<span class="n">math</span><span class="o">.</span><span class="n">ceil</span><span class="p">(</span><span class="n">max_output_len_range</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">/</span> <span class="n">tokens_per_block</span><span class="p">),</span>
<span class="n">math</span><span class="o">.</span><span class="n">ceil</span><span class="p">(</span><span class="n">max_output_len_range</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span> <span class="o">/</span> <span class="n">tokens_per_block</span><span class="p">),</span>
<span class="n">math</span><span class="o">.</span><span class="n">ceil</span><span class="p">(</span><span class="n">max_output_len_range</span><span class="p">[</span><span class="mi">2</span><span class="p">]</span> <span class="o">/</span> <span class="n">tokens_per_block</span><span class="p">)</span>
<span class="p">]]</span>
<span class="n">max_blocks_per_seq_range</span> <span class="o">=</span> <span class="p">[[</span>
<span class="n">x</span> <span class="k">for</span> <span class="n">x</span> <span class="ow">in</span> <span class="n">max_blocks_per_seq_range</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="p">]]</span>
<span class="c1"># PagedKV setup for KV cache of cross-attention</span>
<span class="n">max_cross_blocks_per_seq_range</span> <span class="o">=</span> <span class="p">[[</span>
<span class="n">math</span><span class="o">.</span><span class="n">ceil</span><span class="p">(</span><span class="n">encoder_input_len_range</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">/</span> <span class="n">tokens_per_block</span><span class="p">),</span>
<span class="n">math</span><span class="o">.</span><span class="n">ceil</span><span class="p">(</span><span class="n">encoder_input_len_range</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span> <span class="o">/</span> <span class="n">tokens_per_block</span><span class="p">),</span>
<span class="n">math</span><span class="o">.</span><span class="n">ceil</span><span class="p">(</span><span class="n">encoder_input_len_range</span><span class="p">[</span><span class="mi">2</span><span class="p">]</span> <span class="o">/</span> <span class="n">tokens_per_block</span><span class="p">)</span>
<span class="p">]]</span>
<span class="n">max_cross_blocks_per_seq_range</span> <span class="o">=</span> <span class="p">[[</span>
<span class="n">x</span> <span class="k">for</span> <span class="n">x</span> <span class="ow">in</span> <span class="n">max_cross_blocks_per_seq_range</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="p">]]</span>
<span class="n">num_kv_cache_pools</span> <span class="o">=</span> <span class="mi">2</span>
<span class="n">kv_cache_block_offsets</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="sa">f</span><span class="s1">&#39;kv_cache_block_offsets&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="n">num_kv_cache_pools</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;num_kv_cache_pools&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">num_kv_cache_pools</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;kv&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">2</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;max_blocks_per_seq&#39;</span><span class="p">,</span> <span class="n">max_blocks_per_seq_range</span><span class="p">),</span>
<span class="p">]))</span>
<span class="n">host_kv_cache_block_offsets</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="sa">f</span><span class="s1">&#39;host_kv_cache_block_offsets&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="n">num_kv_cache_pools</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;num_kv_cache_pools&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">num_kv_cache_pools</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;kv&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">2</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;max_blocks_per_seq&#39;</span><span class="p">,</span> <span class="n">max_blocks_per_seq_range</span><span class="p">),</span>
<span class="p">]))</span>
<span class="n">host_kv_cache_pool_pointers</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="sa">f</span><span class="s1">&#39;host_kv_cache_pool_pointers&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int64</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="n">num_kv_cache_pools</span><span class="p">,</span> <span class="mi">2</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;num_kv_cache_pools&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">num_kv_cache_pools</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;num_pools&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">2</span><span class="p">]),</span>
<span class="p">]))</span>
<span class="n">host_kv_cache_pool_mapping</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="sa">f</span><span class="s2">&quot;host_kv_cache_pool_mapping&quot;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="n">num_pp_layers</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;pools_mapping&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">num_pp_layers</span><span class="p">]),</span>
<span class="p">]))</span>
<span class="c1"># paged blocks for cross kv</span>
<span class="n">cross_kv_cache_block_offsets</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="sa">f</span><span class="s1">&#39;cross_kv_cache_block_offsets&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="n">num_kv_cache_pools</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;num_kv_cache_pools&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">num_kv_cache_pools</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;kv&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">2</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;max_cross_blocks_per_seq&#39;</span><span class="p">,</span>
<span class="n">max_cross_blocks_per_seq_range</span><span class="p">),</span>
<span class="p">]))</span>
<span class="n">host_cross_kv_cache_block_offsets</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="sa">f</span><span class="s1">&#39;host_cross_kv_cache_block_offsets&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="n">num_kv_cache_pools</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;num_kv_cache_pools&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">num_kv_cache_pools</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;batch_size_beam_width&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;kv&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">2</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;max_cross_blocks_per_seq&#39;</span><span class="p">,</span>
<span class="n">max_cross_blocks_per_seq_range</span><span class="p">),</span>
<span class="p">]))</span>
<span class="n">host_cross_kv_cache_pool_pointers</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="sa">f</span><span class="s1">&#39;host_cross_kv_cache_pool_pointers&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int64</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="n">num_kv_cache_pools</span><span class="p">,</span> <span class="mi">2</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;num_kv_cache_pools&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">num_kv_cache_pools</span><span class="p">]),</span>
<span class="p">(</span><span class="s1">&#39;num_pools&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">2</span><span class="p">]),</span>
<span class="p">]))</span>
<span class="n">host_cross_kv_cache_pool_mapping</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="sa">f</span><span class="s2">&quot;host_cross_kv_cache_pool_mapping&quot;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="n">num_pp_layers</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;pools_mapping&#39;</span><span class="p">,</span> <span class="p">[</span><span class="n">num_pp_layers</span><span class="p">]),</span>
<span class="p">]))</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="n">layers_range</span><span class="p">:</span>
<span class="n">past_key_value</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="kc">None</span><span class="p">)</span>
<span class="n">kv_cache_params</span> <span class="o">=</span> <span class="n">KeyValueCacheParams</span><span class="p">(</span>
<span class="n">past_key_value</span><span class="o">=</span><span class="n">past_key_value</span><span class="p">,</span>
<span class="n">host_past_key_value_lengths</span><span class="o">=</span><span class="n">host_past_key_value_lengths</span><span class="p">,</span>
<span class="n">host_max_attention_window_sizes</span><span class="o">=</span><span class="n">host_max_attention_window_sizes</span><span class="p">,</span>
<span class="n">host_sink_token_length</span><span class="o">=</span><span class="n">host_sink_token_length</span><span class="p">,</span>
<span class="n">cache_indirection</span><span class="o">=</span><span class="n">cache_indirection</span><span class="p">,</span>
<span class="n">kv_cache_block_offsets</span><span class="o">=</span><span class="n">kv_cache_block_offsets</span><span class="p">,</span>
<span class="n">host_kv_cache_block_offsets</span><span class="o">=</span><span class="n">host_kv_cache_block_offsets</span><span class="p">,</span>
<span class="n">host_kv_cache_pool_pointers</span><span class="o">=</span><span class="n">host_kv_cache_pool_pointers</span><span class="p">,</span>
<span class="n">host_kv_cache_pool_mapping</span><span class="o">=</span><span class="n">host_kv_cache_pool_mapping</span><span class="p">,</span>
<span class="n">cross_kv_cache_block_offsets</span><span class="o">=</span><span class="n">cross_kv_cache_block_offsets</span><span class="p">,</span>
<span class="n">host_cross_kv_cache_block_offsets</span><span class="o">=</span>
<span class="n">host_cross_kv_cache_block_offsets</span><span class="p">,</span>
<span class="n">host_cross_kv_cache_pool_pointers</span><span class="o">=</span>
<span class="n">host_cross_kv_cache_pool_pointers</span><span class="p">,</span>
<span class="n">host_cross_kv_cache_pool_mapping</span><span class="o">=</span>
<span class="n">host_cross_kv_cache_pool_mapping</span><span class="p">,</span>
<span class="p">)</span>
<span class="n">attention_params</span> <span class="o">=</span> <span class="n">AttentionParams</span><span class="p">(</span>
<span class="n">sequence_length</span><span class="o">=</span><span class="n">sequence_length</span><span class="p">,</span>
<span class="n">context_lengths</span><span class="o">=</span><span class="n">context_lengths</span><span class="p">,</span>
<span class="n">host_context_lengths</span><span class="o">=</span><span class="n">host_context_lengths</span><span class="p">,</span>
<span class="n">max_context_length</span><span class="o">=</span><span class="n">max_decoder_input_len</span><span class="p">,</span>
<span class="n">host_request_types</span><span class="o">=</span><span class="n">host_request_types</span><span class="p">,</span>
<span class="n">host_runtime_perf_knobs</span><span class="o">=</span><span class="n">host_runtime_perf_knobs</span><span class="p">,</span>
<span class="n">host_context_progress</span><span class="o">=</span><span class="n">host_context_progress</span><span class="p">,</span>
<span class="n">encoder_input_lengths</span><span class="o">=</span><span class="n">encoder_input_lengths</span><span class="p">,</span>
<span class="n">encoder_max_input_length</span><span class="o">=</span><span class="n">encoder_max_input_length</span><span class="p">,</span>
<span class="p">)</span>
<span class="n">cross_kv_cache_gen</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;cross_kv_cache_gen&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">bool</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;boolean&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">1</span><span class="p">]),</span>
<span class="p">]))</span>
<span class="n">cross_kv_reuse</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">num_heads</span> <span class="o">=</span> <span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">num_heads</span> <span class="o">+</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span> <span class="o">-</span>
<span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span>
<span class="n">cross_kv_out_dim</span> <span class="o">=</span> <span class="mi">2</span> <span class="o">*</span> <span class="n">num_kv_heads</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">head_size</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">skip_cross_kv</span><span class="p">:</span>
<span class="k">if</span> <span class="n">remove_input_padding</span><span class="p">:</span>
<span class="n">cross_kv_reuse</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s2">&quot;cross_kv_reuse&quot;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_dtype</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="n">cross_kv_out_dim</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s2">&quot;encoder_num_tokens&quot;</span><span class="p">,</span> <span class="p">[</span><span class="n">encoder_num_tokens_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s2">&quot;encoder_kv_size&quot;</span><span class="p">,</span> <span class="p">[</span><span class="n">cross_kv_out_dim</span><span class="p">]),</span>
<span class="p">]),</span>
<span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">cross_kv_reuse</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span>
<span class="n">name</span><span class="o">=</span><span class="s2">&quot;cross_kv_reuse&quot;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_dtype</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="n">cross_kv_out_dim</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s2">&quot;batch_size_beam_width_encoder&quot;</span><span class="p">,</span> <span class="p">[</span><span class="n">bb_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s2">&quot;encoder_input_len&quot;</span><span class="p">,</span> <span class="p">[</span><span class="n">encoder_input_len_range</span><span class="p">]),</span>
<span class="p">(</span><span class="s2">&quot;encoder_kv_size&quot;</span><span class="p">,</span> <span class="p">[</span><span class="n">cross_kv_out_dim</span><span class="p">]),</span>
<span class="p">]),</span>
<span class="p">)</span>
<span class="n">skip_cross_attn_blocks</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">skip_cross_attn_blocks</span><span class="p">:</span>
<span class="n">skip_cross_attn_blocks</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;skip_cross_attn_blocks&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">bool</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;boolean&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">1</span><span class="p">]),</span>
<span class="p">]))</span>
<span class="n">prompt_embedding_table</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">tasks</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">prompt_vocab_size</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">is_first_pp_rank</span><span class="p">()</span> <span class="ow">and</span> <span class="n">prompt_embedding_table_size</span> <span class="o">&gt;</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">p_embedding_range</span> <span class="o">=</span> <span class="p">[[</span>
<span class="mi">1</span><span class="p">,</span> <span class="n">prompt_embedding_table_size</span> <span class="o">//</span> <span class="mi">2</span><span class="p">,</span> <span class="n">prompt_embedding_table_size</span>
<span class="p">]]</span>
<span class="n">prompt_embedding_table</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;prompt_embedding_table&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">_dtype</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;prompt_embedding_table_size&#39;</span><span class="p">,</span>
<span class="n">p_embedding_range</span><span class="p">),</span>
<span class="p">(</span><span class="s1">&#39;hidden_size&#39;</span><span class="p">,</span>
<span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">hidden_size</span><span class="p">]),</span>
<span class="p">]))</span>
<span class="k">if</span> <span class="n">remove_input_padding</span><span class="p">:</span>
<span class="n">num_tokens_range</span> <span class="o">=</span> <span class="p">[</span>
<span class="mi">1</span><span class="p">,</span>
<span class="p">(</span><span class="n">max_decoder_input_len</span> <span class="o">*</span> <span class="n">max_batch_size</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)</span> <span class="o">//</span> <span class="mi">2</span><span class="p">,</span>
<span class="n">max_decoder_input_len</span> <span class="o">*</span> <span class="n">max_batch_size</span><span class="p">,</span>
<span class="p">]</span>
<span class="n">tasks</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;tasks&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;decoder_num_tokens&#39;</span><span class="p">,</span>
<span class="p">[</span><span class="n">decoder_num_tokens_range</span><span class="p">]),</span>
<span class="p">]))</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">tasks</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;tasks&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([</span>
<span class="p">(</span><span class="s1">&#39;batch_size&#39;</span><span class="p">,</span> <span class="n">bs_range</span><span class="p">),</span>
<span class="p">(</span><span class="s1">&#39;broadcast_dim&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">1</span><span class="p">]),</span>
<span class="p">]))</span>
<span class="n">prompt_vocab_size</span> <span class="o">=</span> <span class="n">Tensor</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;prompt_vocab_size&#39;</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">trt</span><span class="o">.</span><span class="n">int32</span><span class="p">,</span>
<span class="n">shape</span><span class="o">=</span><span class="p">[</span><span class="mi">1</span><span class="p">],</span>
<span class="n">dim_range</span><span class="o">=</span><span class="n">OrderedDict</span><span class="p">([(</span><span class="s1">&#39;size&#39;</span><span class="p">,</span> <span class="p">[</span><span class="mi">1</span><span class="p">])]))</span>
<span class="n">result</span> <span class="o">=</span> <span class="p">{</span>
<span class="s1">&#39;decoder_input_ids&#39;</span><span class="p">:</span> <span class="n">input_ids</span><span class="p">,</span>
<span class="s1">&#39;encoder_output&#39;</span><span class="p">:</span> <span class="n">encoder_output</span><span class="p">,</span>
<span class="s1">&#39;use_cache&#39;</span><span class="p">:</span> <span class="kc">True</span><span class="p">,</span>
<span class="s1">&#39;attention_mask_params&#39;</span><span class="p">:</span> <span class="n">attention_mask_params</span><span class="p">,</span>
<span class="s1">&#39;last_token_ids&#39;</span><span class="p">:</span> <span class="n">last_token_ids</span><span class="p">,</span>
<span class="s1">&#39;kv_cache_params&#39;</span><span class="p">:</span> <span class="n">kv_cache_params</span><span class="p">,</span>
<span class="s1">&#39;attention_params&#39;</span><span class="p">:</span> <span class="n">attention_params</span><span class="p">,</span>
<span class="s1">&#39;hidden_states&#39;</span><span class="p">:</span> <span class="n">hidden_states</span><span class="p">,</span>
<span class="s1">&#39;lora_params&#39;</span><span class="p">:</span> <span class="n">lora_params</span><span class="p">,</span>
<span class="s1">&#39;cross_kv_cache_gen&#39;</span><span class="p">:</span> <span class="n">cross_kv_cache_gen</span><span class="p">,</span>
<span class="s1">&#39;cross_kv_reuse&#39;</span><span class="p">:</span> <span class="n">cross_kv_reuse</span><span class="p">,</span>
<span class="s1">&#39;prompt_embedding_table&#39;</span><span class="p">:</span> <span class="n">prompt_embedding_table</span><span class="p">,</span>
<span class="s1">&#39;prompt_tasks&#39;</span><span class="p">:</span> <span class="n">tasks</span><span class="p">,</span>
<span class="s1">&#39;prompt_vocab_size&#39;</span><span class="p">:</span> <span class="n">prompt_vocab_size</span><span class="p">,</span>
<span class="s1">&#39;skip_cross_attn_blocks&#39;</span><span class="p">:</span> <span class="n">skip_cross_attn_blocks</span><span class="p">,</span>
<span class="p">}</span>
<span class="k">return</span> <span class="n">result</span></div>
<div class="viewcode-block" id="MLLaMAModel.use_lora">
<a class="viewcode-back" href="../../../../python-api/tensorrt_llm.models.html#tensorrt_llm.models.MLLaMAModel.use_lora">[docs]</a>
<span class="k">def</span> <span class="nf">use_lora</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">lora_config</span><span class="p">:</span> <span class="n">LoraConfig</span><span class="p">):</span>
<span class="n">use_lora</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">lora_config</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">trtllm_modules_to_hf_modules</span><span class="p">)</span></div>
<div class="viewcode-block" id="MLLaMAModel.precompute_relative_attention_bias">
<a class="viewcode-back" href="../../../../python-api/tensorrt_llm.models.html#tensorrt_llm.models.MLLaMAModel.precompute_relative_attention_bias">[docs]</a>
<span class="k">def</span> <span class="nf">precompute_relative_attention_bias</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">build_config</span><span class="p">):</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">relative_attention</span> <span class="ow">and</span> <span class="ow">not</span> <span class="bp">self</span><span class="o">.</span><span class="n">use_implicit_relative_attention</span><span class="p">:</span>
<span class="n">relative_attention_bias_builder</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ops</span><span class="o">.</span><span class="n">tensorrt_llm</span><span class="o">.</span><span class="n">relative_attention_bias</span>
<span class="n">rel_attn_precomputed</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">zeros</span><span class="p">(</span>
<span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">num_attention_heads</span> <span class="o">//</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="n">build_config</span><span class="o">.</span><span class="n">max_seq_len</span> <span class="o">+</span> <span class="mi">1</span><span class="p">,</span> <span class="n">build_config</span><span class="o">.</span><span class="n">max_seq_len</span> <span class="o">+</span> <span class="mi">1</span><span class="p">),</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">str_dtype_to_torch</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">dtype</span><span class="p">),</span>
<span class="n">device</span><span class="o">=</span><span class="s1">&#39;cuda&#39;</span><span class="p">)</span>
<span class="n">rel_attn_table</span> <span class="o">=</span> <span class="n">numpy_to_torch</span><span class="p">(</span>
<span class="bp">self</span><span class="o">.</span><span class="n">rel_attn_table</span><span class="o">.</span><span class="n">raw_value</span><span class="p">)</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="s1">&#39;cuda&#39;</span><span class="p">)</span>
<span class="n">relative_attention_bias_builder</span><span class="p">(</span>
<span class="n">rel_attn_precomputed</span><span class="p">,</span>
<span class="n">rel_attn_table</span><span class="p">,</span>
<span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">num_attention_heads</span> <span class="o">//</span> <span class="bp">self</span><span class="o">.</span><span class="n">mapping</span><span class="o">.</span><span class="n">tp_size</span><span class="p">,</span>
<span class="n">build_config</span><span class="o">.</span><span class="n">max_seq_len</span><span class="p">,</span>
<span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">num_buckets</span><span class="p">,</span>
<span class="kc">False</span><span class="p">,</span>
<span class="bp">self</span><span class="o">.</span><span class="n">config</span><span class="o">.</span><span class="n">max_distance</span><span class="p">,</span>
<span class="p">)</span>
<span class="k">for</span> <span class="n">layer_idx</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">num_layers</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">decoder_layers</span><span class="p">[</span>
<span class="n">layer_idx</span><span class="p">]</span><span class="o">.</span><span class="n">self_attention</span><span class="o">.</span><span class="n">set_rel_attn_table</span><span class="p">(</span>
<span class="n">build_config</span><span class="o">.</span><span class="n">max_seq_len</span><span class="p">,</span> <span class="n">rel_attn_precomputed</span><span class="p">)</span></div>
<div class="viewcode-block" id="MLLaMAModel.from_hugging_face">
<a class="viewcode-back" href="../../../../python-api/tensorrt_llm.models.html#tensorrt_llm.models.MLLaMAModel.from_hugging_face">[docs]</a>
<span class="nd">@classmethod</span>
<span class="k">def</span> <span class="nf">from_hugging_face</span><span class="p">(</span>
<span class="bp">cls</span><span class="p">,</span>
<span class="n">hf_model_or_dir</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="s1">&#39;transformers.PreTrainedModel&#39;</span><span class="p">],</span>
<span class="n">dtype</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s1">&#39;auto&#39;</span><span class="p">,</span>
<span class="n">mapping</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Mapping</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="n">quant_config</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">QuantConfig</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
<span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
<span class="w"> </span><span class="sd">&#39;&#39;&#39; Create a MLLaMAModel object from give parameters</span>
<span class="sd"> &#39;&#39;&#39;</span>
<span class="kn">import</span> <span class="nn">transformers</span>
<span class="n">kwargs</span><span class="o">.</span><span class="n">pop</span><span class="p">(</span><span class="s1">&#39;load_by_shard&#39;</span><span class="p">,</span> <span class="kc">False</span><span class="p">)</span>
<span class="n">kwargs</span><span class="o">.</span><span class="n">pop</span><span class="p">(</span><span class="s1">&#39;load_model_on_cpu&#39;</span><span class="p">,</span> <span class="kc">False</span><span class="p">)</span>
<span class="n">quant_ckpt_path</span> <span class="o">=</span> <span class="n">kwargs</span><span class="o">.</span><span class="n">pop</span><span class="p">(</span><span class="s1">&#39;quant_ckpt_path&#39;</span><span class="p">,</span> <span class="kc">None</span><span class="p">)</span>
<span class="k">assert</span> <span class="n">hf_model_or_dir</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span>
<span class="n">use_preloading</span> <span class="o">=</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">hf_model_or_dir</span><span class="p">,</span>
<span class="n">transformers</span><span class="o">.</span><span class="n">PreTrainedModel</span><span class="p">)</span>
<span class="k">if</span> <span class="n">use_preloading</span><span class="p">:</span>
<span class="n">hf_model</span> <span class="o">=</span> <span class="n">hf_model_or_dir</span>
<span class="n">hf_config_or_dir</span> <span class="o">=</span> <span class="n">hf_model</span><span class="o">.</span><span class="n">config</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">hf_model_dir</span> <span class="o">=</span> <span class="n">hf_model_or_dir</span>
<span class="n">hf_config_or_dir</span> <span class="o">=</span> <span class="n">hf_model_or_dir</span>
<span class="n">config</span> <span class="o">=</span> <span class="n">MLLaMAConfig</span><span class="o">.</span><span class="n">from_hugging_face</span><span class="p">(</span><span class="n">hf_config_or_dir</span><span class="p">,</span>
<span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span>
<span class="n">mapping</span><span class="o">=</span><span class="n">mapping</span><span class="p">,</span>
<span class="n">quant_config</span><span class="o">=</span><span class="n">quant_config</span><span class="p">,</span>
<span class="o">**</span><span class="n">kwargs</span><span class="p">)</span>
<span class="n">custom_dict</span> <span class="o">=</span> <span class="p">{</span>
<span class="s2">&quot;lm_head&quot;</span><span class="p">:</span> <span class="s2">&quot;language_model.lm_head&quot;</span><span class="p">,</span>
<span class="s2">&quot;ln_f&quot;</span><span class="p">:</span> <span class="s2">&quot;language_model.model.norm&quot;</span><span class="p">,</span>
<span class="s2">&quot;decoder_layers&quot;</span><span class="p">:</span> <span class="s2">&quot;language_model.model.layers&quot;</span><span class="p">,</span>
<span class="s2">&quot;self_attention&quot;</span><span class="p">:</span> <span class="s2">&quot;self_attn&quot;</span><span class="p">,</span>
<span class="s2">&quot;cross_attention&quot;</span><span class="p">:</span> <span class="s2">&quot;cross_attn&quot;</span><span class="p">,</span>
<span class="s2">&quot;vocab_embedding&quot;</span><span class="p">:</span> <span class="s2">&quot;language_model.model.embed_tokens&quot;</span><span class="p">,</span>
<span class="s2">&quot;gate_attn&quot;</span><span class="p">:</span> <span class="s2">&quot;cross_attn_attn_gate&quot;</span><span class="p">,</span>
<span class="s2">&quot;gate_ffwd&quot;</span><span class="p">:</span> <span class="s2">&quot;cross_attn_mlp_gate&quot;</span><span class="p">,</span>
<span class="s2">&quot;q_layernorm&quot;</span><span class="p">:</span> <span class="s2">&quot;q_norm&quot;</span><span class="p">,</span>
<span class="s2">&quot;k_layernorm&quot;</span><span class="p">:</span> <span class="s2">&quot;k_norm&quot;</span><span class="p">,</span>
<span class="p">}</span>
<span class="k">if</span> <span class="n">quant_ckpt_path</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">hf_model_dir</span> <span class="o">=</span> <span class="n">quant_ckpt_path</span>
<span class="n">loader</span> <span class="o">=</span> <span class="n">ModelWeightsLoader</span><span class="p">(</span><span class="n">hf_model_dir</span><span class="p">,</span> <span class="n">custom_dict</span><span class="p">)</span>
<span class="n">model</span> <span class="o">=</span> <span class="bp">cls</span><span class="p">(</span><span class="n">config</span><span class="p">)</span>
<span class="n">loader</span><span class="o">.</span><span class="n">generate_tllm_weights</span><span class="p">(</span><span class="n">model</span><span class="p">)</span>
<span class="k">return</span> <span class="n">model</span></div>
</div>
</pre></div>
</div>
</div>
<footer>
<hr/>
<div role="contentinfo">
<jinja2.runtime.BlockReference object at 0x7f1ac71bddc0>
<div class="footer">
<p>
Copyright © 2024 NVIDIA Corporation
</p>
<p>
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/privacy-policy/" target="_blank" rel="noopener"
data-cms-ai="0">Privacy Policy</a> |
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/privacy-center/" target="_blank" rel="noopener"
data-cms-ai="0">Manage My Privacy</a> |
<a class="Link" href="https://www.nvidia.com/en-us/preferences/start/" target="_blank" rel="noopener"
data-cms-ai="0">Do Not Sell or Share My Data</a> |
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/terms-of-service/" target="_blank"
rel="noopener" data-cms-ai="0">Terms of Service</a> |
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/accessibility/" target="_blank" rel="noopener"
data-cms-ai="0">Accessibility</a> |
<a class="Link" href="https://www.nvidia.com/en-us/about-nvidia/company-policies/" target="_blank"
rel="noopener" data-cms-ai="0">Corporate Policies</a> |
<a class="Link" href="https://www.nvidia.com/en-us/product-security/" target="_blank" rel="noopener"
data-cms-ai="0">Product Security</a> |
<a class="Link" href="https://www.nvidia.com/en-us/contact/" target="_blank" rel="noopener"
data-cms-ai="0">Contact</a>
</p>
</div>
</div>
</footer>
</div>
</div>
</section>
</div>
<script>
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
</body>
</html>
Reference in New Issue View Git Blame Copy Permalink