This PR moves MTPSampler and derivatives to use the universal seq_slot indexing for sampling.
This is the last piece of the puzzle: After this, all of the samplers will use this format.
See: 6ee94c7
Signed-off-by: Netanel Haber <nhaber@nvidia.com>
58a8a8f - these changes were previously merged to main here.
6aef149 - the changes were temporarily reverted in main, due to a significant perf regression in models using the TorchSampler (observed by @byshiue).
This PR is meant to re-merge these changes along with a fix to prevent the regression.
The first commit of this PR is actually just the reverted revert - filter it out of the changes to see previously unmerged changes.
Signed-off-by: Netanel Haber <nhaber@nvidia.com>
The seq_len of 4096 will cause some unknown CUDA illegal memory access issue if run with some other tests consecutively.
Put a saturated upper bound for any sequence length larger than it.
- Adds a new Python custom op (fp8_block_scale_moe_runner) and a FP8BlockScaleMoERunner class for autotuning.
- Updates C++ MoE and batched GEMM kernels to accept a configIndex for workspace sizing and execution.
- Extends the unit test to run both autotuned and non-autotuned code paths.
Signed-off-by: Dom Brown <3886319+DomBrown@users.noreply.github.com>
Signed-off-by: Barry Kang <43644113+Barry-Delaney@users.noreply.github.com>
Signed-off-by: Omer Ullman Argov <118735753+omera-nv@users.noreply.github.com>
Co-authored-by: Barry Kang <43644113+Barry-Delaney@users.noreply.github.com>
Fix AutoTuner warmup request generating.
* The current warmup phase creates one request, which is insufficient for the warmup to cover the max_num_tokens. Revise the warmup phase to a batch of requests to cover the max_num_tokens to eliminate potential fallback cases.
Refactor AutoTuner API and reduce host overhead.
Refine (min, opt, max) values of optimization profile setup for get_valid_tactics to achieve the correct canImplement definition.
* Refine cache key assembly process to reduce host overhead and simplify API.
* Fix lru_cache usage to reduce host overhead.
* Move tuning config initialization as a one-time object in tunable runner to reduce host overhead.
Improve tuning config readability.
* Use dataclass to define tuning config.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
Signed-off-by: Hao Lu <14827759+hlu1@users.noreply.github.com@users.noreply.github.com>
Co-authored-by: Hao Lu <14827759+hlu1@users.noreply.github.com@users.noreply.github.com>
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
Co-authored-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
Signed-off-by: Hao Lu <14827759+hlu1@users.noreply.github.com@users.noreply.github.com>
Co-authored-by: Hao Lu <14827759+hlu1@users.noreply.github.com@users.noreply.github.com>
Signed-off-by: Hao Lu <14827759+hlu1@users.noreply.github.com@users.noreply.github.com>
Co-authored-by: Hao Lu <14827759+hlu1@users.noreply.github.com@users.noreply.github.com>
* refine shutdown signal of PyExecutor
Signed-off-by: QI JUN <22017000+QiJune@users.noreply.github.com>
* clean
Signed-off-by: QI JUN <22017000+QiJune@users.noreply.github.com>
* fix ci
Signed-off-by: QI JUN <22017000+QiJune@users.noreply.github.com>
* fix ci
Signed-off-by: QI JUN <22017000+QiJune@users.noreply.github.com>
---------
Signed-off-by: QI JUN <22017000+QiJune@users.noreply.github.com>
* Fixed typos in the scaffolding README.MD
Signed-off-by: Anton <44649959+amemov@users.noreply.github.com>
* Fixed links for 'More examples' and 'Contribute Guide'
Signed-off-by: Anton <44649959+amemov@users.noreply.github.com>
---------
Signed-off-by: Anton <44649959+amemov@users.noreply.github.com>
* refactor: CreateNewDecoderRequests
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Consolidate request generation in CreateNewDecoderRequests
- Removed the GenerateRequestOptions class and integrated its functionality into CreateNewDecoderRequests.
- Updated the constructor of CreateNewDecoderRequests to accept parameters for speculative decoding and normalization options.
- Modified the operator() method to handle request generation directly, improving code organization and reducing redundancy.
- Cleaned up associated includes and references throughout the codebase.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Simplify request handling in CreateNewDecoderRequests
- Removed the generateRequestOptions method and integrated its logic directly into the operator() method.
- Updated the request generation process to improve clarity and reduce redundancy.
- Adjusted the return type to streamline the handling of batch slots, decoder requests, and sampling configurations.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Enhance createDecoderRequests method in CreateNewDecoderRequests
- Updated the createDecoderRequests method to include additional parameters for decoder state and CUDA streams, improving flexibility in request handling.
- Removed redundant request generation logic from the operator() method, streamlining the process.
- Adjusted the newRequest method to utilize the updated decoder request structure, enhancing clarity and maintainability.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Use MedusaBuffers instead of RuntimeBuffers in CreateNewDecoderRequests
- Updated references from RuntimeBuffers to MedusaBuffers across the CreateNewDecoderRequests class and its methods, enhancing clarity in buffer management.
- Adjusted method signatures and internal logic to accommodate the new MedusaBuffers type, ensuring compatibility with existing functionality.
- Cleaned up unnecessary includes and improved code organization for better maintainability.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Update CreateNewDecoderRequests to use DecoderState and CudaStream parameters
- Modified method signatures in CreateNewDecoderRequests to replace GptDecoderBatched with runtime::decoder::DecoderState and added a separate CudaStream for the decoder.
- Adjusted the implementation of the operator() method to accommodate the new parameters, enhancing flexibility in request handling.
- Updated associated bindings in the pybind11 interface to reflect the changes in method signatures, ensuring consistency across the codebase.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Update TRTLLMSampler to use refactored create_new_decoder_requests
- Updated the sampler.py to reflect changes in the request handling logic, replacing generate_request_options with create_new_decoder_requests for improved clarity and consistency.
- Updated bindings and method signatures for decoder stream handling.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Update gptDecoderBatchedTest to use CreateNewDecoderRequests::newRequest
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
---------
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* MoE TRTLLM backend for Qwen3
Signed-off-by: Anthony Chang <anchengc@nvidia.com>
* add extra moe_backend to test
Signed-off-by: Anthony Chang <anchengc@nvidia.com>
* address comments
Signed-off-by: Anthony Chang <anchengc@nvidia.com>
* conditionally compile kernels on newer archs
Signed-off-by: Anthony Chang <anchengc@nvidia.com>
* missing positional arg
Signed-off-by: Anthony Chang <anchengc@nvidia.com>
* Update the routing kernels
Signed-off-by: Christina Zhang <christinaz@nvidia.com>
* Revise usage of TLLM_LOG_ERROR
Signed-off-by: Christina Zhang <christinaz@nvidia.com>
* Add unit test for Qwen3 moe (trtllm_gen backend)
Signed-off-by: Christina Zhang <christinaz@nvidia.com>
* improve weight processing speed of moe_backend=TRTLLM; roughly 2x
Signed-off-by: Anthony Chang <anchengc@nvidia.com>
* tidy and minor fix
Signed-off-by: Anthony Chang <anchengc@nvidia.com>
* temporarily disable accuracy test that has known issue
Signed-off-by: Anthony Chang <anchengc@nvidia.com>
---------
Signed-off-by: Anthony Chang <anchengc@nvidia.com>
Signed-off-by: Christina Zhang <christinaz@nvidia.com>
Co-authored-by: Christina Zhang <christinaz@nvidia.com>
* fix bug of qwen3 fp4 workflow with EP
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
* fix bug of qwen3_moe with ep
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
---------
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
* added a restricted pcikler and depickler in a sepparate serialization function.
Signed-off-by: coldwaterq@users.noreply.github.com <coldwaterq@users.noreply.github.com>
* updated IPC to remove approved classes, removed the serialization function because it didn't work for all objects that made debugging harder, added tests.
Signed-off-by: coldwaterq@users.noreply.github.com <coldwaterq@users.noreply.github.com>
* removed LLM arg and moved class registration to a serialization module function. Also added missing classes to approved list.
Signed-off-by: coldwaterq <coldwaterq@users.noreply.github.com>
* cleaned up a couple files to reduce conflicts with main.
Signed-off-by: coldwaterq <coldwaterq@users.noreply.github.com>
* fix unit tests
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
* reorder BASE_ZMQ_CLASSES list alphabetically
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
* fix tests and move LogitsProcessor registration to base class
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
* revert changes to import log of tensorrt_llm._torch.models
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
* added comments to explain why BASE_ZMQ_CLASSES has to be passed into spawned child processes
Signed-off-by: coldwaterq <coldwaterq@users.noreply.github.com>
* fix tests and move LogitsProcessor registration to base class
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
* additional comments for multiprocess approved list sync
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
* add dataclass from tests
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
---------
Signed-off-by: coldwaterq@users.noreply.github.com <coldwaterq@users.noreply.github.com>
Signed-off-by: coldwaterq <coldwaterq@users.noreply.github.com>
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
Co-authored-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
* add bidirectional support and fix EarlyStopDecoder unsqueeze to be compatible with LogitsStorage
Signed-off-by: Rohan Varma <rohanv@nvidia.com>
* run pre-commit
Signed-off-by: Rohan Varma <rohanv@nvidia.com>
* instead of bidirectional flag use ModelConfig.is_generation
Signed-off-by: Rohan Varma <rohanv@nvidia.com>
* fix unit test to extract logits from correct dim
Signed-off-by: Rohan Varma <rohanv@nvidia.com>
---------
Signed-off-by: Rohan Varma <rohanv@nvidia.com>
* Fix TRTLLMSampler.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Added type hint.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
---------
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* first commit of cpp moe loadbalance code
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
* add python bindings for moe load balance
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
* add python wrapper, ut and bug fixes
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
* add binding for layerId and update binding test
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
* add host tensor sharing and ut
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
---------
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
* [AutoDeploy] HF factory improvements
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
* improve monkey-patches and add unit tests
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
---------
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
This PR adds a customized allreduce to TensorRT-LLM. The new allreduce is used for communication on PCIe-based GPUs via low-precision quantization, which can accelerate the PCIe allreduce process.
Signed-off-by: Hui Kang <hkang@nvidia.com>
Co-authored-by: Hui Kang <hkang@nvidia.com>
* Add llama4 disagg accuracy tests
Signed-off-by: Iman Tabrizian <10105175+tabrizian@users.noreply.github.com>
* Make it async and add GSM8K benchmark
Signed-off-by: Iman Tabrizian <10105175+tabrizian@users.noreply.github.com>
---------
Signed-off-by: Iman Tabrizian <10105175+tabrizian@users.noreply.github.com>
* Deduce default max_tokens for trtllm-serve
Signed-off-by: Pengyun Lin <81065165+LinPoly@users.noreply.github.com>
* Improve executor_config.max_seq_len assignment in TRT workflow
Signed-off-by: Pengyun Lin <81065165+LinPoly@users.noreply.github.com>
* Enhance error message
Signed-off-by: Pengyun Lin <81065165+LinPoly@users.noreply.github.com>
* Add deduced max_tokens test
Signed-off-by: Pengyun Lin <81065165+LinPoly@users.noreply.github.com>
---------
Signed-off-by: Pengyun Lin <81065165+LinPoly@users.noreply.github.com>
chore: restore symmetry of worker start/shutdown
chore: fix return type of cal_max_tokens
chore: type some more return values
fix: free resources before re-claiming
Signed-off-by: ixlmar <206748156+ixlmar@users.noreply.github.com>
* refactor: Copy sequence lengths once in decoder setup
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Update DecoderInputBuffers to remove duplicated buffers
- Renamed and reorganized buffer variables in decoderBuffers.h and decoderBuffers.cpp for better readability.
- Adjusted references in generateRequestOptions.cpp to align with the new buffer structure.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Move getEmbeddingBias to anonymous namespace
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Filter context requests
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: GenerateRequestOptions using more fine-grained functions
- Added a new method `createDecoderRequests` to encapsulate the logic for creating decoder requests from finished context requests.
- Updated the `operator()` method to utilize the new method, improving code clarity and maintainability.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Update TRTLLMDecoder
- Updated the `generate_request_options` call.
- Updated the `make_decoding_batch_input_output` call.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Remove const where we modify input buffers
- Changed `DecoderInputBuffers` parameters from const references to non-const references in multiple functions to allow modifications.
- Updated related function calls to ensure compatibility with the new parameter types.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* fixup! refactor: Copy sequence lengths once in decoder setup
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
---------
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* add docstring to summarize current rope support
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* minor: replace call_method, adjust inserting order of cos_sin_cache calculation node
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* add unit test for triton rope and ds rope
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* update rope matcher to match DS RoPE, add custom op for reference, add unit test case
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* cache cos[pos_idx].unsqueeze and sin[pos_idxs].unsqueeze
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* minor doc update
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* separate pattern matching and optimization for explicit and complex rope + minor updates
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* clean rope impl in repo
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* replace fused_flattened_mla_with_cache's rope impl with torch_apply_rope_with_qk_interleaving, update unit test
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* minor
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* separate layout infer and transpose to a new transformation
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* update rope_with_explicit_freqs and rope_with_input_interleaved to expose unsqueeze_dim and support match_rope_layout, add unit tests
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* solve merge conflict in transform.py, need to fix optimize_rope with cuda graph capture
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* minor clean up after rebase
Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>
* fix pre-commit
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* support map to bnsd layout and infer unsqueeze_dim from op
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* fix cos/sin not the same across prompts in the same batch issue when mapping to flashinfer op
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* fix for unit test
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* fix custom op input/output node ordering issue for DeepSeek V3 rope
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* clean code
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* minor
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* move flattening of cos_sin_cache to the graph, update flashinfer op docstring and test
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* debug transform unit test failure
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
---------
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>
Signed-off-by: Fridah-nv <201670829+Fridah-nv@users.noreply.github.com>
* update matcher to match expert compute first, then extract other args with LCA
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* support 3D and 2D input in torch.ops.moe.trtllm_fused_moe
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* update custom ops to support 3D and 2D inputs
Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>
* update deepseek patch
Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>
---------
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* Add test case for kv memory estimation
* Dump running log into file and parse kv cache memory size from file
* Set bigger peak memory size for mixed percision case and test_ptp_quickstart_advanced_eagle3 case
* Revert change to usage of fraction
* use context manager to guard temp files
Signed-off-by: Hui Gao <huig@nvidia.com>
This PR adds a customized allreduce to TensorRT-LLM. The new allreduce is used for communication on PCIe-based GPUs via low-precision quantization, which can accelerate the PCIe allreduce process.
Signed-off-by: Hui Kang <hkang@nvidia.com>
Co-authored-by: Hui Kang <hkang@nvidia.com>
Support DeepSeek-R1 W4A8 on Hopper
Co-authored-by: Barry Kang <43644113+Barry-Delaney@users.noreply.github.com>
Co-authored-by: Jiang Shao <91270701+StudyingShao@users.noreply.github.com>
Signed-off-by: Barry Kang <43644113+Barry-Delaney@users.noreply.github.com>
* Add chunking to PyT heuristic.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
* Cast tokens and batch size to ints.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
---------
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
This issue is found for tp=ep=8 on the multi-node machine due to the inconsistent PP sizes.
* Reform the workspace allocation implementation to avoid the list-out-of-range issues.
* Disable min_latency_mode under the multi-node scenario to avoid the illegal memory access issue.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
Prefetching safetensors files so that they are stored in the system file
cache. This significantly speeds up the model weight loading for the
very first run after entering the docker container.
This is beneficial because model weight loading is done layer-by-layer,
which means reading from the safetensors chunk-by-chunk, and that cannot
utilize the internet bandwidth very well, assuming that these files are
stored in some network drives. Instead, loading the whole files in bulk
can achieve higher internet bandwidth utilization.
When running with world_size>1, all ranks collaboratedly prefetch these
files.
In theory, we should add heuristics to decide whether to prefetch the
files or not, but that is beyond the scope of this commit.
For example, when the CPU memory is small, doing prefetching may result
in file cache thrashing, resulting in slower weight loading time.
Signed-off-by: Po-Han Huang <pohanh@nvidia.com>
* feat: Add heuristic for GroupRMSNorm kernel selection.
Implements a logistic regression model to dynamically select between:
- GroupRMSNormBaseKernel: Allocates warps proportional to sum of dimensions
(better SM occupancy in most cases)
- GroupRMSNormLargeBatch: Allocates warps proportional to max dimension
(better block scheduling in large batch scenarios)
Selection heuristic considers batch size, allocated warps, and scheduling
efficiency on the current GPU architecture. Models for Compute Capability
9.x and 10.x are trained base on nsys kernel runtime data.
The default kernel selection is the base kernel.
The python operator group_rms_norm will use the heuristic by default.
User can pick to use the base or large batch kernels as well.
Signed-off-by: Simeng Liu <simengl@nvidia.com>
* Address the comments.
Signed-off-by: Simeng Liu <simengl@nvidia.com>
---------
Signed-off-by: Simeng Liu <simengl@nvidia.com>
* Set cuda graph max batch size.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
* Set padding.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
---------
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
* fix add_dummy_requests.
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
* add max_seq_len to eagle3 test and fix add_dummy_requests.
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
* fix prompt_len in add_dummy_requests.
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
* add prepare_resource condition in add_dummy_requests.
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
* add some description of token_nums to add_dummy_requests and fix token_nums in torch compile warmup.
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
* fix available_tokens.
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
---------
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
* fix relaxed acceptance to support enable this feature in context phase.
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
* fix sample_and_accept_draft_tokens unit test.
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
---------
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
* Fallback to NCCL for various patterns when input size is large.
Move the previous implementation to cpp side.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Revising.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
---------
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Properly get decoding mode according to same logic as cpp.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Cross reference getDecodingMode implementations in pytorch - cpp.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Better bindings for DecodingMode.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Revert to version in main.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Fix.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Revert configuration.py.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
---------
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* disable overlap in encoder
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* feat: invokeGatherBatch
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* feat: overlap same batch
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* chore: add enableTrtOverlap to ExecutorConfig
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* disable overlap for beam search and spec decode
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* skip overlap tests with beam search or speculative decoding
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* moveFinishedContextRequestsToGeneration and skip unfinished requests in updateRequests
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* enable overlap in GptChunkedLongContextTests
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* feat: Enable overlap in gptManagerBenchmark
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* feat: Improve early exit
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Use OptionalRef for newOutputTokens tensor
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* feat: Add overlap scheduling support to TRTLLMDecoder
- Updated TRTLLMDecoder to accept an `enable_overlap_scheduler` parameter.
- Modified the decoder's internal logic to utilize the overlap scheduling feature.
- Adjusted the sequence lengths handling to ensure compatibility with the new scheduling approach.
- Enhanced unit tests to include cases for the overlap scheduler with the TRTLLMDecoder.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* fix: allNewTokens in PP
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
---------
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* Instantiate decoder early to have better mem estimation.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Improve mem estimation by instantiating decoder earlier.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
---------
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Fix AllReduce kernel hang issue when both tp and pp are enabled.
Allocate one workspace for each pp rank to avoid potential race.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* update waive list
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
---------
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* feat: Add rename_weights_with_regex function for dynamic weight key renaming
Introduced a new utility function to rename weight keys in a dictionary based on regex pattern matching. This allows for flexible mapping of keys from Hugging Face naming conventions to TRT-LLM naming conventions, enhancing model compatibility and usability.
Signed-off-by: qixiang-99 <203170375+qixiang-99@users.noreply.github.com>
* feat: Implement SiglipVisionModel and related components
Added the SiglipVisionModel along with its associated classes, including SiglipAttention, SiglipEncoderLayer, and SiglipEncoder.
Additionally, a new test suite for the SiglipVisionModel has been created to ensure compatibility with Hugging Face outputs.
Currently SiglipVisionModel support batch size larger than one. Also, inputs and outputs shape are same with the HF for compatibility.
Signed-off-by: qixiang-99 <203170375+qixiang-99@users.noreply.github.com>
* feat: Add CLIPVisionModel and associated components
Introduced the CLIPVisionModel along with its related classes, including CLIPAttention, CLIPEncoderLayer, CLIPEncoder, and CLIPVisionTransformer. This implementation aligns with Hugging Face's CLIP architecture, ensuring compatibility in input and output shapes.
Signed-off-by: qixiang-99 <203170375+qixiang-99@users.noreply.github.com>
* feat: Enhance CLIPVisionModel with attention metadata preparation and unit tests
Updated the CLIPVisionModel to include a method for preparing attention metadata, simplifying the model's usage. Additionally, added a comprehensive unit test suite for the CLIPVisionModel, ensuring compatibility with Hugging Face outputs and validating model performance across various scenarios.
Signed-off-by: qixiang-99 <203170375+qixiang-99@users.noreply.github.com>
* feat: Refactor SiglipVisionModel with attention metadata preparation and update unit tests
Enhanced the SiglipVisionModel by adding a method to prepare attention metadata, streamlining its usage. Updated unit tests to validate model performance and compatibility with Hugging Face outputs, including adjustments to the configuration and test scenarios.
Signed-off-by: qixiang-99 <203170375+qixiang-99@users.noreply.github.com>
* refactor: Remove unused rotary_emb parameter from CLIP and Siglip attention classes
Eliminated the rotary_emb parameter from the CLIPAttention and SiglipAttention classes to streamline the code. Updated unit tests to reflect changes in the model configurations, including clarifications in the default configurations sourced from Hugging Face.
Signed-off-by: qixiang-99 <203170375+qixiang-99@users.noreply.github.com>
* feat: Integrate CLIPVisionModel into LlavaNextInputProcessor and enhance weight loading
Added CLIPVisionModel to the LlavaNextInputProcessor for improved vision processing. Updated the model loading mechanism to ensure compatibility with the new vision model and added attention metadata preparation. Removed debug print statements from weight renaming function for cleaner code.
Signed-off-by: qixiang-99 <203170375+qixiang-99@users.noreply.github.com>
* refactor: Remove unused max_position_embeddings from CLIPAttention and update Siglip classes to use CLIP components
Removed the unused max_position_embeddings variable from the CLIPAttention class. Updated the Siglip classes to utilize CLIP components, specifically replacing SiglipEncoder and SiglipAttention with their CLIP counterparts, streamlining the codebase and enhancing consistency across models.
Signed-off-by: qixiang-99 <203170375+qixiang-99@users.noreply.github.com>
* refactor: Consolidate weight loading logic into a shared implementation
Refactored the weight loading process across CLIP and Siglip models by using a new utility function, _load_weights_impl, to streamline the loading mechanism. This change enhances code maintainability and reduces redundancy in weight handling, ensuring consistent behavior across different model architectures.
Signed-off-by: qixiang-99 <203170375+qixiang-99@users.noreply.github.com>
* refactor: Simplify output handling in CLIP and Siglip models by removing output_hidden_states parameter
Removed the output_hidden_states parameter from the CLIPEncoder and SiglipVisionTransformer classes, streamlining the output handling process. Updated the corresponding unit tests to reflect these changes and ensure compatibility with the new output structure.
Signed-off-by: qixiang-99 <203170375+qixiang-99@users.noreply.github.com>
* feat: Enhance LlavaNextInputProcessor with dynamic model loading and memory optimization
Updated the LlavaNextInputProcessor to support dynamic model loading from local paths or Hugging Face, improving memory efficiency by partially loading the model components. Integrated the LlavaNextMultiModalProjector and adjusted weight loading to ensure compatibility with the new architecture.
Signed-off-by: qixiang-99 <203170375+qixiang-99@users.noreply.github.com>
---------
Signed-off-by: qixiang-99 <203170375+qixiang-99@users.noreply.github.com>
Co-authored-by: Haohang Huang <31998628+symphonylyh@users.noreply.github.com>
* Move all casters to customCasters.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Use customCasters in all bindings.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Added customCasters to userbuffers.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
---------
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* feat: Add group_rms_norm kernel to normalize multiple inputs in a single operator.
Previously, the RMSNorm implementation only supported a single input tensor. With group_rms_norm, multiple tensors can be normalized together:
```python
input_a, input_b, ... = group_rms_norm([input_a, input_b, ...])
```
All input tensors must share the same batch dimension. The kernel partitions work by dynamically assigning warp groups proportional to the last dimension of each input, improving launch efficiency and reducing overhead.
This MR provides two implementations:
GroupRMSNormKernel: Optimized for small-to-medium batch sizes
GroupRMSNormKernelLargeBatch: Contains additional optimizations for large batch sizes
Both kernels are currently exposed as custom PyTorch ops. A future MR will implement heuristic-based kernel selection and expose a unified interface.
Signed-off-by: Simeng Liu <simengl@nvidia.com>
* Resolve comments and fix typo with IS_FLASHINFER_AVAILABLE
Signed-off-by: Simeng Liu <simengl@nvidia.com>
---------
Signed-off-by: Simeng Liu <simengl@nvidia.com>
When input size is larger than the max workspace size, we shall fallback to NCCL + corresponding pre/post function to ensure the functionality of AllReduce.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* support lp in pytorch backend
Signed-off-by: Erin Ho <14718778+hchings@users.noreply.github.com>
* fix tp
Signed-off-by: Erin Ho <14718778+hchings@users.noreply.github.com>
---------
Signed-off-by: Erin Ho <14718778+hchings@users.noreply.github.com>
* support return logprob in llmapi
Signed-off-by: Erin Ho <14718778+hchings@users.noreply.github.com>
update and add test
Signed-off-by: Erin Ho <14718778+hchings@users.noreply.github.com>
stability test
Signed-off-by: Erin Ho <14718778+hchings@users.noreply.github.com>
* revert removal of old flag
Signed-off-by: Erin Ho <erinh@nvidia.com>
Signed-off-by: Erin Ho <14718778+hchings@users.noreply.github.com>
---------
Signed-off-by: Erin Ho <14718778+hchings@users.noreply.github.com>
Signed-off-by: Erin Ho <erinh@nvidia.com>
* add qwen3 dense model pytorch backend support, initial commit
solve the results error issue
add qwen3 moe model pytorch backend support
reformat the code
* perf - use flash_infer rmsnorm for qwen3
* feat - support qwen3 moe rmsnorm
* Put the computation of Q and K norm (in attn) into a single CUDA stream, and get a 5% - 8% throughput improvement on Qwen3 4B and Qwen3 - moe 30B - A3B.
* Put the computation of Q and K norm (in attn) into a single CUDA stream, and get a 5% - 8% throughput improvement on Qwen3 4B and Qwen3 - moe 30B - A3B. -- Forgot to update all modifications.
* fix bugs of running qwen3 public models and fp8 models
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
* fix bugs due to rebase
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
* fix bugs captured by pre-commi
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
* fix bug of attention
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
---------
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
Co-authored-by: Keddy Jin <jin.gq@aliyun.com>
Co-authored-by: Jiying Dong <87510204+dongjiyingdjy@users.noreply.github.com>
Co-authored-by: shao <shao@nvidia.com>
* Add a new param to LlmRequest and Request to natively support mm
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* update comment
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Update tests to match the new LlmRequest constructor parameters
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Modify unitTest and modify mm_embeding's dict name in llama4
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Fix based on comments
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Fix comment
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Fix LlmRequest initialization in kvCacheManagerTest
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Clean up code for promt_tuning_config
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Clean up prompt_tuning_config in GenerationRequest
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
---------
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
Co-authored-by: Haohang Huang <31998628+symphonylyh@users.noreply.github.com>
* Move world options to a different group for clarity.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
* Add eos_id option.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
---------
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
* Replace deepseek_allreduce op with the new unified allreduce op and moe_allreduce op.
* Minor revision of moe_allreduce op argument names.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
(1) match quant exclude modules names to TRTLLM names
(2) No need for any special weight loading for quantization scales weights (#3891)
Signed-off-by: Tomer Asida <57313761+tomeras91@users.noreply.github.com>
* add parallel_q_b_proj_and_concat
Signed-off-by: junliu <65336694+hello-11@users.noreply.github.com>
* code cleanup
Signed-off-by: junliu <65336694+hello-11@users.noreply.github.com>
* one gemm/concat and then split the latent_cache and pass them separately to context/gen
Signed-off-by: junliu <65336694+hello-11@users.noreply.github.com>
---------
Signed-off-by: junliu <65336694+hello-11@users.noreply.github.com>
test: add test cases for 0.19 release (#3608)
* fix test name
* add quickstart test for nemotron-ultra
* add rcca multi-node test case for deepseek-v3
* add rcca info
---------
squash (#3642)
fix: nvbugs/5187237: fix deterministic mode crash (#3448)
* nvbugs/5187237 nvbugs/5112075: fix deterministic mode error
* remove waive
* Revert "remove waive"
This reverts commit 0bf5486d19906d692bfb7a6262333c296b0087ac.
* revert ar fusion
---------
update fp8 doc (#3647)
tests: change qa perf test to trtllm-bench (#3619)
fix: FP8 quantized lm_head (NvBug 5214229) (#3567)
infra: Add PR approval protection for the release branch (#3634)
fix: nvbugs/5231298: pytorch allreduce issue (#3673)
Fix: nvbugs/5222698 variable not defined (#3630)
* Fix: nvbugs/5222698 variable not defined
* Tidy code
---------
test:sync waives.txt from main branch by disabling test_perf/gpt_350m-cppmanager case (#3685)
test:restore fp8 kv cache testing for L0 (#3671)
doc: Update DeepSeek perf docs (#3693)
* Update DeepSeek perf docs
* update
* Apply suggestions from code review
---------
tests: waive test_llm_multi_node (#3664)
fix: update test_user_buffers_mm_add_prologue atol (#3711)
Fix: cherry-pick hmac encryption from main branch (#3635)
* security fix cherry-pick changes from main
* fix hmac in remote mpi session (#3649)
---------
Un-waive DS-V3-Lite tests. (#3621)
fix: FP8 kv accuracy (#3675)
* fix FP8 kv accuracy
* update doc
---------
Fix script options for engines. (#3622)
unwaive multi-node test (#3721)
chore : Split more tests out of gpt tests (#3524) (#3674)
doc:add torch examples link into torch backend documentation (#3749)
test: Get Eagle tests working (#3593) (#3722)
Waive L0 test (#3756)
waive failed case in perf test, change default max_batch_size to 512 and write config.json to output log (#3656)
Update ds v3 parameters in stress test. (#3676)
waive gemma on L20 (#3766)
https://nvbugs/5141291: Fix convert.py script for Qwen model. (#3758)
Include Qwen2VLDecoderLayer in the smooth_qwen2_model function.
fix: PP4 fixes and cleanup (#3688)
remove benchmark test list (#3643)
skip disagg deepseek test if sm!=90 (#3720)
test: skip failed cases on B200 (#3710)
* add skip condition to tests
* fix error
---------
test: [nvbug: 5234494] skip_pre_ada for fp8 cases (#3718)
* skip_pre_ada for fp8 cases
* update
* update after rebase
---------
add know issue to deepseek doc. (#3800)
Fix ModelOpt Mixtral AWQ OOM (#3714) (#3761)
Waive L0 tests (#3826)
fix: Reduce memory usage in fused moe op associated with AutoTuning and fix moe fallback issue. (#3793)
* Reduce memory usage in fused moe op associated with AutoTuning.
* Replace pre-defined bucket size strategy with a generating function based on the tune_max_num_tokens.
* Add free_memory logic of workspace in min_latency_mode fused moe path.
* Fix fused_moe fallback issue. (#3652)
min_latency_mode is only set to False during warmup phase. Thus when it becomes true during inference, all tactics fall back to the default one and thus cause perf regression.
---------
[doc] Better document for Draft-Target-Model (DTM) speculative decoding (#3797)
Fix pre-commit
Fix again
Address some review comments for the MI
Signed-off-by: Dom Brown <3886319+DomBrown@users.noreply.github.com>
Co-authored-by: Zhanrui Sun <184402041+ZhanruiSunCh@users.noreply.github.com>
* Update gen tps calculation.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
* Add back output speed for comparison.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
* Fix issue with f-string.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
* Fix some spacing.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
* Replace output speed with per-request genphase tput.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
* Add gen TPS breakdown.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
* Update some tagging.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
---------
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
Signed-off-by: Hao Lu <14827759+hlu1@users.noreply.github.com@users.noreply.github.com>
Co-authored-by: Hao Lu <14827759+hlu1@users.noreply.github.com@users.noreply.github.com>
* fix bug of create cuda stream as default parameter which will be initialized during importing
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
* add torch.cuda.Stream() for the leader node
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
* fix pre-commit issue
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
---------
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
* add MNNVL memory mapping support
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
* add more MPI environment for trtllm-llmapi-launch
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
* add MoE communication and prepare kernels
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
* add MNNVL AlltoAll support for DeepSeekV3
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
* add output dump for throughput benchmark
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
* support dynamic kernel launch grid
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
* address review comments
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
* address review comments #2
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
---------
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
* Use updateDecoderBuffers in python decoder.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Fix synchronize in trtllm decoder.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Enable by default.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Use guided_decoder to setup seqslots and free them.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Use always decode_async and update_requests.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Update decoder buffers.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Fix speculative decoding tests.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Send new_tensors_host instead of assuming dict.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Make default False in enable_trtllm_decoder.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Partially fix mtp, partially fix py_executor.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Update request states before sending disagg ctx cache.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Fix disagg test for torch decoder.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Make isend_tensor_list and recv_tensor_list for sending the tensors_host.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Formatting.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Fix rebase.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Add disagg serving case to guided decoder.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Get overlap scheduling to work.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Update cutlass to main.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Update after rebasing.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Formatting.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Update to use decode async and update requests.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Properly pass information to update_requests
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Formatting.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Make disaggregated serving a step closer to working.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Fix rebase.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Fix rebase and format.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Copy new device tokens more pythonic.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Restore MTP add dummy reqs.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Add ordereddict import to py_executor.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Formatting.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Added seq slot manager. Add test.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Use transmission for single tensor except when list of tensors is received.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Add TRTLLMDecoder allocation to estimate max kv cache tokens.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Add stream synchronization
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Formatting.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Make memory calculation of decoder adapt to the chosen decoder. Recognize decoder option passed in executorconfig. Make overlap scheduler test run on TinyLlama.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Format
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Add decoder creation to estimate max kv.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Formatting.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Update submodule UCXX inline with main.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
---------
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* add passing E2E LoRA flow
Signed-off-by: Shahar Mor <smor@nvidia.com>
* add experimental feature
Signed-off-by: Shahar Mor <smor@nvidia.com>
* fix llma_args definition
Signed-off-by: Shahar Mor <smor@nvidia.com>
* decreased manually size of max loras to address OOM
Signed-off-by: Shahar Mor <smor@nvidia.com>
---------
Signed-off-by: Shahar Mor <smor@nvidia.com>
* generalizing cudagraph to multiple dynamic inputs
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
* fix for failing test
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
---------
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
* Rewrite unit test for unified allreduce op. Removing the legacy unit test.
* Revise formats, fusion_op bindings. Put all tensors as optional inputs.
* Move the MoeAllreduceOp to a separate custom op.
* Move all the fusion patterns to the new version of the AllReduce fusion kernel. Remove the AllReduce strategy config. Revise the AllReduce strategies and fusion pattern definitions.
* Add more TODOs, fixing minor bugs, and remove legacy code.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Fix hang bug when KV cache is low
Signed-off-by: Iman Tabrizian <itabrizian@nvidia.com>
* Review comments
Signed-off-by: Iman Tabrizian <itabrizian@nvidia.com>
* Fix attentiondp typo
Signed-off-by: Iman Tabrizian <itabrizian@nvidia.com>
* Add CI test for this case
Signed-off-by: Iman Tabrizian <itabrizian@nvidia.com>
* fix: Fix the insertion order for responder futures
Signed-off-by: Iman Tabrizian <10105175+tabrizian@users.noreply.github.com>
* fix: Fix disagg CPP
Signed-off-by: Iman Tabrizian <10105175+tabrizian@users.noreply.github.com>
---------
Signed-off-by: Iman Tabrizian <itabrizian@nvidia.com>
Signed-off-by: Iman Tabrizian <10105175+tabrizian@users.noreply.github.com>
* Feat: Offload ptable to cpu if enable_chunk_context
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Feat: offload ptable to cpu for chunk context mode
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Fix and add comment
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Update Readme for multimodal and add a new param mm_embedding_offloading
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* fix: Correct prompt table offloading condition in PromptTuningBuffers
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Clean up the code
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Add commits to explain copy from cpu <-> gpu using pinned memory
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Fix namings based on comments
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Fix format based on precommit
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
* Modify --mm_embedding_offloading flag
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
---------
Signed-off-by: Kate Cheng <yunhsuanc@nvidia.com>
Co-authored-by: Haohang Huang <31998628+symphonylyh@users.noreply.github.com>
* feat: Integrate GPUDirect Storage (GDS) into Executor API
Squash of several dev commits
Signed-off-by: Dom Brown <3886319+DomBrown@users.noreply.github.com>
* feat: adding multimodal (only image for now) support in trtllm-bench
Signed-off-by: Rakib Hasan <rhasan@nvidia.com>
* fix: add in load_dataset() calls to maintain the v2.19.2 behavior
Signed-off-by: Rakib Hasan <rhasan@nvidia.com>
* re-adding prompt_token_ids and using that for prompt_len
Signed-off-by: Rakib Hasan <rhasan@nvidia.com>
* updating the datasets version in examples as well
Signed-off-by: Rakib Hasan <rhasan@nvidia.com>
* api changes are not needed
Signed-off-by: Rakib Hasan <rhasan@nvidia.com>
* moving datasets requirement and removing a missed api change
Signed-off-by: Rakib Hasan <rhasan@nvidia.com>
* addressing review comments
Signed-off-by: Rakib Hasan <rhasan@nvidia.com>
* refactoring the quickstart example
Signed-off-by: Rakib Hasan <rhasan@nvidia.com>
---------
Signed-off-by: Rakib Hasan <rhasan@nvidia.com>
min_latency_mode is only set to False during warmup phase. Thus when it becomes true during inference, all tactics fall back to the default one and thus cause perf regression.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* implement variable window attention by breaking the block manager into window block managers per window size
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* revert isCyclic to be true if the min attention window is reached, not per window size
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* add explanatory comment to mCyclicThreshold
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* load correct gemma config
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* don't shadow inputLength in addSequence - it should remain the function scope input length between window size loop iterations
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* fix KVCacheManagerVariableWindowAttentionWithReuseTest for multiple window block managers
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* if TYPE_CHECKING
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* set temp_attention_window_inputs to None explicitly
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* set temp_attention_window_inputs to None explicitly
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* pass dtype as well
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* test_gemma variable sliding window attention
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* allot a fraction of primary/secondaryBlocks to different window size heaps, depending on the window size's total contribution to the kvcache size (i.e., including all layers)
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* remove || mEnableBlockReuse which erroneously triggers beamsearch code for cyclic variable attention window code
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* turn off request delaying for MaxUtil
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* make comments better
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* windowSizesTotalSum using std::accumulate
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* fix error handling of forwardAsync - forwardAsync catch-all catch cleanup code that runs terminateRequest can also fail and must be caught
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* fix comments
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* remove assert that kills disagg tests, since it isn't necessary
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* fix corrupted expression: 'isNewTask && (peftCacheManager ?' -> '(isNewTask && peftCacheManager) ?' which caused boolean algebra. Main is correct
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* add Gemma3 to SUPPORTED_HF_ARCHITECTURES
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* support Gemma3
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* finally fix test_gemma - always spread at least {} into generate_summary_cmd, never None
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* finally fix test_gemma - always spread at least {} into generate_summary_cmd, never None
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* fix kvfactor field for deepseek
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* fix comment
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* fix gemma-3 entries in testlist to include vswa
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* only quantize gemma2 VSWA
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
remove misleading comment
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
fix test_gemma
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* fix test_gemma
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* fix test_gemma
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* in sendRequestInfo, fromOldAllocatedBlockIds->fromOldAllocatedBlockIds, like in main
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
* fix: disable KV cache reuse if using attention sink (#3021)
* fix: disable KV cache reuse if using attention sink
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* fix: disable KV cache reuse if sink bubble
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* add comment
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
---------
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
---------
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
Co-authored-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* added files for nemotron-h
Signed-off-by: Luis Vega <lvega@nvidia.com>
* use try/except to import RMSNorm
Signed-off-by: Luis Vega <lvega@nvidia.com>
---------
Signed-off-by: Luis Vega <lvega@nvidia.com>
Co-authored-by: QI JUN <22017000+QiJune@users.noreply.github.com>
* hmac initial implementation to encrypt worker and proxy queue
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
* set different hmac key for each pair of server/client queue
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
* fix comments
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
* fix style
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
---------
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
* [Infra][TRTLLM-4063] - Branch out for the TRT-LLM v0.18.0 release
Signed-off-by: Zhanrui Sun <zhanruis@nvidia.com>
(cherry picked from commit de90312020e51c22ba5e75b3502c7ee90c059265)
* [Infra][TRTLLM-3652] - Update dependencies to TRT 10.9 / CUDA 12.8.1 / DLFW 25.03(Internal)
Signed-off-by: Yiqing Yan <yiqingy@nvidia.com>
(cherry picked from commit 58db1340ef7db22f1910f878d220a92be5b830d1)
* [None][Doc] - Update docs for v0.18.0
Signed-off-by: Yanchao Lu <yanchaol@nvidia.com>
(cherry picked from commit d23e75bc95619ce3b116213d55319272888e0c88)
* [Infra] - Fix or WAR issues in the package sanity check stages
Signed-off-by: Yanchao Lu <yanchaol@nvidia.com>
(cherry picked from commit e874e2b127515c52ba10c8df1cc2631627f74ffe)
* [https://nvbugs/5173454] [https://nvbugs/5173432] [https://nvbugs/5175863] fix chatglm tokenizer and tmp model path
Signed-off-by: Yuki Huang <yukih@nvidia.com>
(cherry picked from commit 731811d4e182d70a66193d646152cb71dfafe83a)
* cherry-pick 'test: Updat cluster and multi node test lists and trtllm-bench' test to fix perf drop issue
Signed-off-by: Ruodi Lu <ruodil@nvidia.com>
(cherry picked from commit 5214616283fbc15ae98871a1d84c78d8e1f2e6e8)
* Revert "Merge branch 'user/yukih/fix_5173454_5173432' into 'release/0.18'"
Signed-off-by: Yanchao Lu <yanchaol@nvidia.com>
(cherry picked from commit 8d34831cb2b81ee2dfa8021b68e7158b33789a5f)
* [Infra]Restrict setuptools version to avoid sasb pip install issue
Signed-off-by: Emma Qiao <qqiao@nvidia.com>
(cherry picked from commit 1e60ad29e0dafec0e295bedb5d89b716a02a707c)
* [https://nvbugs/5173454] [https://nvbugs/5173432] [https://nvbugs/5175863] fix chatglm tokenizer and tmp model path
Signed-off-by: Yuki Huang <yukih@nvidia.com>
(cherry picked from commit 3ed8164e5bfea1d5aa2039b5408439fd6cf59dac)
* WAR for bug 5173448
Signed-off-by: Thor Johnsen <tjohnsen@nvidia.com>
(cherry picked from commit b6528b2ba15322b6c6a4c81a8b74c04d4973de4f)
* [Infra][TRTLLM-3652] - Update dependencies to CUDA 12.8.1 / DLFW 25.03
Signed-off-by: Yiqing Yan <yiqingy@nvidia.com>
(cherry picked from commit 6560983d132d9d257ee15849664eb055e94adaa9)
* [Docs] - Doc changes for v0.18.0
Signed-off-by: Yanchao Lu <yanchaol@nvidia.com>
(cherry picked from commit 26769b61218a947c8f9d070f73b63d576fcc20c4)
* [Doc] - Doc change for v0.18.0
Signed-off-by: Yanchao Lu <yanchaol@nvidia.com>
(cherry picked from commit 4b3b5ed6bfbc2300e3775fe75456083faad7b235)
* [Infra] update version to 0.18.1
Signed-off-by: Zhanrui Sun <zhanruis@nvidia.com>
(cherry picked from commit 59e8326c75639275837d34de8e140358737a3365)
* Add back nemotron file.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Fix recurrentgemma reqs.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Adding WAR for bug 5173448.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Formatting.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Remove duplicated file.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Update examples/prompt_lookup/requirements.txt
Co-authored-by: Zhanrui Sun <184402041+ZhanruiSunCh@users.noreply.github.com>
Signed-off-by: Daniel Cámpora <961215+dcampora@users.noreply.github.com>
* Remove glm-4-9b from model dir in chatglm test.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Remove indent change.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Apply suggestions from code review
Co-authored-by: Yanchao Lu <yanchaol@nvidia.com>
Signed-off-by: Daniel Cámpora <961215+dcampora@users.noreply.github.com>
* Apply suggestions from code review
Co-authored-by: Yanchao Lu <yanchaol@nvidia.com>
Signed-off-by: Daniel Cámpora <961215+dcampora@users.noreply.github.com>
* Revert changes on l0_test.groovy.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Update dev images
Co-authored-by: Zhanrui Sun <184402041+ZhanruiSunCh@users.noreply.github.com>
Signed-off-by: Yanchao Lu <yanchaol@nvidia.com>
* Remove duplicated import.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Fix custom op
Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>
* Fix flashinfer & vanilla backend
Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>
* Skip problematic case.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
* Skip problematic test_moe_w4a8_1_14336_4096_8_bfloat16_True_False case.
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
---------
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
Signed-off-by: Daniel Cámpora <961215+dcampora@users.noreply.github.com>
Signed-off-by: Yanchao Lu <yanchaol@nvidia.com>
Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>
Co-authored-by: Zhanrui Sun <zhanruis@nvidia.com>
Co-authored-by: Yiqing Yan <yiqingy@nvidia.com>
Co-authored-by: Yanchao Lu <yanchaol@nvidia.com>
Co-authored-by: Yuki Huang <yukih@nvidia.com>
Co-authored-by: Ruodi Lu <ruodil@nvidia.com>
Co-authored-by: Emma Qiao <qqiao@nvidia.com>
Co-authored-by: Thor Johnsen <tjohnsen@nvidia.com>
Co-authored-by: Zhanrui Sun <184402041+ZhanruiSunCh@users.noreply.github.com>
Co-authored-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>
Co-authored-by: Tao Li @ NVIDIA <tali@nvidia.com>
* chore: unify pp_layers helpers
Fix assumptions about equal number of layers per PP rank
in prepare_attention_inputs
Signed-off-by: Aurelien Chartier <2567591+achartier@users.noreply.github.com>
* apply a tenative fix to moe bypass kernel update
* Pass none to disable final stage in moe
Co-authored-by: hlu1 <14827759+hlu1@users.noreply.github.com>
Signed-off-by: Chang Liu <lc9114@gmail.com>
---------
Signed-off-by: Chang Liu <lc9114@gmail.com>
Co-authored-by: hlu1 <14827759+hlu1@users.noreply.github.com>
* fix: Fixing issue with first gen token being returned twice with streaming
Signed-off-by: Patrice Castonguay <55748270+pcastonguay@users.noreply.github.com>
* Fixing not_expectring_strings in test
Signed-off-by: Patrice Castonguay <55748270+pcastonguay@users.noreply.github.com>
---------
Signed-off-by: Patrice Castonguay <55748270+pcastonguay@users.noreply.github.com>
Co-authored-by: QI JUN <22017000+QiJune@users.noreply.github.com>
* fix: add kv memory size per token of draft model to calculate max number
of tokens of kv cache
Signed-off-by: Hui Gao
* Fix code to get model_config of draft model
Signed-off-by: Hui Gao
---------
Signed-off-by: Hui Gao
* refactor: remove cumLogProbs and logProbs from DecoderBuffers
- Eliminated cumLogProbs and logProbs from DecoderBuffers, streamlining the buffer management.
- Updated related code in decoderBuffers.cpp and bindings.cpp to reflect these changes, ensuring that only host pointers are used for log probabilities.
These modifications enhance code clarity and maintainability by reducing redundancy in buffer management.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: streamline sequence length handling in GptDecoderBatched and StatefulGptDecoderBatched
- Updated GptDecoderBatched to directly use output.sequenceLengths for lengths assignment, removing unnecessary reshaping.
- Adjusted StatefulGptDecoderBatched to ensure sequence lengths are correctly shaped based on actual batch size and max beam width.
These changes enhance clarity and maintainability in the decoding process.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: integrate DecoderState for sequence length management in decoding process
- Updated DecoderBuffers to remove direct handling of sequence lengths, now utilizing DecoderState for this purpose.
- Adjusted MakeDecodingBatchInputOutput to accept DecoderState, enhancing clarity in the decoding input/output management.
- Refactored GptDecoderBatched and StatefulGptDecoderBatched to streamline sequence length handling, ensuring consistency across the decoding workflow.
refactor: update SlotDecoderBuffers to manage sequence lengths directly
- Introduced sequenceLengths and sequenceLengthsHost to SlotDecoderBuffers for better management of sequence lengths.
- Refactored asyncSend and recv methods to utilize the new sequenceLengths member, enhancing clarity and reducing redundancy.
- Updated TrtGptModelInflightBatching to align with the new structure, ensuring consistent handling of sequence lengths across the decoding process.
These changes improve maintainability and streamline the decoding workflow.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* refactor: Delegate to asyncSend method in SlotDecoderBuffers
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
---------
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* add test to map flashinfer rope op with triton custom rope ops and pytorch rope in fused_mha
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* add rope matcher and unit tests
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* capture cos and sin from graph
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* revert fuse_mha op change
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* minor update to address comment and remove redundant unit test
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* move view and transpose into graph nodes and update unit test to test custom op directly
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* move view into custom op, update bfs with bound, update custom op return type to be half precision
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* custom op update to support 3D input
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* handle bnsd and bsnd format, update tests, handle 3D cos/sin input to the custom op
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* add llama4 rope test, update custom op with is_neox flag
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* add llama4 style rope to matcher and update unit test
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* separate into two transformations
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* fix when num_head != num_kv_head; add support for cached position_ids and cos_sin_cache in graph; update unit tests
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* minor update, cache locally and propagate meta info of qk nodes
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* minor: fix cos_sin_cache not float
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* minor: move cache into matcher
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
---------
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
* One of the tactic is not supported during dispatch.
* final_hidden_states should be unpacked if it is not min_latency_mode.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* feat: Add NVFP4 UB pattern optimization pass in torch compile
* Add an additional flag for UB fp4 pattern to avoid inverse the scale
* Add NVFP4 related UB patterns
Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
* Update atol, some points fails for B200 umbriel.
Signed-off-by: liji-nv <59594262+liji-nv@users.noreply.github.com>
---------
Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
Signed-off-by: liji-nv <59594262+liji-nv@users.noreply.github.com>
* Rename nvsmall to nemotron NAS
* Revert nvsmall to nemotron_nas rename in paths in tests that access llm_models_root/nvsmall/tests
* Add NemotronNAS to pytorch supported models table
Signed-off-by: Amit Zuker <203509407+amitz-nv@users.noreply.github.com>
* Add nested aliases for Llama 4
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
* Fix missed alias.
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
---------
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
Co-authored-by: Yan Chunwei <328693+Superjomn@users.noreply.github.com>
* fix: broadcast embeddings input when using speculative decoding
Signed-off-by: Maximiliano Levi <maxilevi77@gmail.com>
* fix: use shape tensor instead of tuple
Signed-off-by: Maximiliano Levi <maxilevi77@gmail.com>
* fix: comment
Signed-off-by: Maximiliano Levi <maxilevi77@gmail.com>
---------
Signed-off-by: Maximiliano Levi <maxilevi77@gmail.com>
Co-authored-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
* Instead of allocating UserBuffers at beginning of runtime, UB buffers
are now managed with global allocator. The allocator will dynamically
assign free UB buffer or allocate new buffer for torch tensor. It makes
userbuffers easier to use.
* In common usecase, the Userbuffers will be allocated correctly during
warm up stage. There is no dynamic allocation during inference.
* UB fusion pattern is rewroten using the new UB Allocator. It contains
following passes:
1. Fuse Quant with allreduce, replace with UB impl, and insert a
copy_to_userbuffers. Currently the normal allreduce still does not
support FP8 quant. So this need to be done in UB pass
2. Convert all supported allreduce with UB and insert copy_to_userbuffers.
3. Fuse op before ar with the copy_to_userbuffers. So the op directly
writes to the userbuffer
4. Remove userbuffers finalize if the output is connect to another UB
allreduce.
Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
* Several optimizations and fixings on the Autotuner.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Apply the new Python side Autotuner on current linear for nvFP4 data type.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Apply the new Python side Autotuner on MoE op
* Remove routers from cache key to improve inference perf
* Prevent unnecessary code profiling. Use do_preparation keyword to select which part should be executed during before evaluating any tactic.
* Remove try-catch inside moe profiling process.
* Move default tactic -1 to 0 transforms in cpp runner.
* Revise relavant tests.
* Predefined the bucketizing strategy for fused_moe
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Add specific_profile support for AutoTuner to bypass the standard cache search process for perf optimization
* Add specific_profile for moe
* Add specific profile for linear
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Fixing and revising according to reviewer's suggestions.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Use lru_cache for inference pref optimization.
* Revert gen_custom_cache_key feature
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Replace runner with runner id to achieve a serializable cache.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Code clean up and minor fixings.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Move all tunable runners and custom ops into torch_custom_ops.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* Treat min_latency_mode as a independent dynamic tensor. Modify get_valid_tactics to suit for it.
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
---------
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
* feat: Add option to run disaggregated serving without ctx servers, to benchmark gen only
Signed-off-by: Patrice Castonguay <55748270+pcastonguay@users.noreply.github.com>
* Fixing comment in sanity check
Signed-off-by: Patrice Castonguay <55748270+pcastonguay@users.noreply.github.com>
---------
Signed-off-by: Patrice Castonguay <55748270+pcastonguay@users.noreply.github.com>
* fp8 kv + bf16 ctx MLA + fp8 gen MLA
Use BF16 for context MLA.
mFP8GenerationMLA and mFP8ContextFMHA shouldn't be enabled together.
Allow mSM==90 for mFP8GenerationMLA==true.
For FMHA, dataTypeKv should be FP8.
For FP8 MLA generation, the output is still in BF16.
Refine debug info for FMHA kernel metadata.
Use inputType, outputType, SM together to hash kernel list.
Add FP8 MLA generation FMHA kernel.
Special WAR of NUM_COMPUTE_GROUPS for MLA generation kernel.
Separate the implementation of fused_multihead_attention_v2.h to CPP and print some debug info if checkIfKernelExist fails.
Refine debug info in fused_multihead_attention_v2.cpp
Correct FP8 MLA metadata.
New kernel provided by Yuxin, which outputs BF16.
smem size is not set correctly, which will lead to illegal mem access.
Yuxin fixed the error in FMHA MLA kernel: previously the BF16 isn't correctly written: some parts are repeatedly written, while some others are untouched.
There are two bmm1 scales that should be set correctly.
New kernel generated by Yuxin.
Modificatiosn to common/attentionOp for FP8 MLA on Hopper using FMHA.
Not necessary. If mFP8GenerationMLA, is_fp8_out is false, so mFP8ContextFMHA is false.
Skip a check in fmhaDispatcher.
Modifications in fmhaRunner:
- Debug dump.
- if (!isFP8GenerationMLA) skips a lot of flag setting.
- TMA descriptor modification for qo (by Yuxin).
Cleanup debug output.
Clean up o tma descriptor modifications.
Signed-off-by: Bo Li <bobboli0202@gmail.com>
* Resolve conflicts.
Signed-off-by: Bo Li <bobboli0202@gmail.com>
* Apply the patch of FP8 FlashMLA and resolve conflicts.
Signed-off-by: Bo Li <bobboli0202@gmail.com>
* Fix compilation error.
Signed-off-by: Bo Li <bobboli0202@gmail.com>
* Fix compile error.
Signed-off-by: Bo Li <bobboli0202@gmail.com>
* pick blackwell support
Signed-off-by: Dylan Chen <191843203+DylanChen-NV@users.noreply.github.com>
* Add copyright notice to fused_multihead_attention_v2.cpp.
Signed-off-by: Bo Li <bobboli0202@gmail.com>
* Add license.
Signed-off-by: Bo Li <bobboli0202@gmail.com>
* Add missing license.
Signed-off-by: Bo Li <bobboli0202@gmail.com>
* Exclude building flashMLA kernels under sm90.
Signed-off-by: Bo Li <bobboli0202@gmail.com>
* Revert "Exclude building flashMLA kernels under sm90."
This reverts commit f0c859d459.
Signed-off-by: Bo Li <bobboli0202@gmail.com>
* Use macro to skip compiling FlashMLA for non sm90 targets.
Signed-off-by: Bo Li <bobboli0202@gmail.com>
---------
Signed-off-by: Bo Li <bobboli0202@gmail.com>
Signed-off-by: Dylan Chen <191843203+DylanChen-NV@users.noreply.github.com>
Co-authored-by: Dylan Chen <ziqingc@nvidia.com>
Co-authored-by: Dylan Chen <191843203+DylanChen-NV@users.noreply.github.com>
Co-authored-by: QI JUN <22017000+QiJune@users.noreply.github.com>
* refactor: Expose DecoderState via bindings and integrate in TRTLLMDecoder
- Introduced a new `DecoderState` class in the C++ bindings, encapsulating key functionalities for managing decoding state.
- Adjusted the Python `TRTLLMDecoder` to access properties from `decoder_state`, ensuring consistency and clarity in the decoding process.
These changes streamline the decoder's architecture and enhance maintainability.
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* chore: Remove unused new_tokens from DecoderState bindings
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
---------
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
* init trtllm attn no cache
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* fix: fix the seq_len issue and attn metadata prepare for qwen reward model test
fix: fix minor bugs after rebase
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* refactor: remove unnecessary debug logs and clean up commented code
refactor: update max_seq_len documentation and remove max_seq_len for decoder model contructor in PyTorchModelEngine
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* refactor: update calculate_ref_result function to accept tensor inputs and mask type, enhance test_attention_no_cache to support FULL and CAUSAL masks
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* refactor: remove unused BERT attention metadata conversion method and add type assertion for no cache attention in PyTorchModelEngine
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* refactor: remove use_kv_cache parameter from attention function and related classes, update documentation for KV cache handling
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* refactor: implement setAttentionMaskType method for better mask type handling and remove unused conversion function
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* refactor: streamline KV cache handling by replacing direct member access with useKVCache method and simplify token per block assignment
remove Debug code.
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* refactor: Resolve comments for Python code
Simplify no cache attention metadata preparation and streamline related attributes in TrtllmAttentionMetadata
Removed the private method for converting to no cache attention metadata and integrated its logic into the prepare method. Updated the test for BERT sequence classification to reflect these changes and ensure proper handling of attention metadata.
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* docs: Add is_dummy_attention field to attention metadata for simulation operations
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* refactor: add KVCacheParams to attention backend interface and import relevant metadata classes
Updated the attention backend interface to include KVCacheParams and imported TrtllmAttentionMetadata and VanillaAttentionMetadata in model_engine.py for enhanced functionality.
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* fix: fix rebase format issue
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* fix: extend attention mask type handling in MHARunnerFixedParams
Added support for additional attention mask types (BIDIRECTIONAL, BIDIRECTIONALGLM, BLOCKSPARSE) in the MHARunnerFixedParams structure to fix the mapping issue between ContextAttentionMaskType and AttentionMaskType
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* fix: enhance attention mask type handling in TllmGenFmhaRunnerParams
Updated the setAttentionMaskType method to include a switch-case structure for better handling of attention mask types, ensuring proper mapping and error handling for invalid types.
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
---------
Signed-off-by: Qixiang Lin <qixiangl@nvidia.com>
* fix acceptance rate of pytorch workflow.
* revert the RequestOutput API change.
---------
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
