# SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved. # SPDX-License-Identifier: Apache-2.0 # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import argparse import copy import functools import json import os import time import traceback from collections import defaultdict from concurrent.futures import ThreadPoolExecutor, as_completed from typing import Dict, Optional, Tuple import numpy as np import safetensors import torch import torch.nn as nn from tqdm import tqdm from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer from transformers.pytorch_utils import Conv1D import tensorrt_llm from tensorrt_llm.mapping import Mapping from tensorrt_llm.models.convert_utils import weight_only_quantize_dict from tensorrt_llm.quantization import QuantAlgo def parse_arguments(): parser = argparse.ArgumentParser() parser.add_argument('--model_dir', type=str, default=None) parser.add_argument('--quant_ckpt_path', type=str, default=None) parser.add_argument('--tp_size', type=int, default=1) parser.add_argument('--pp_size', type=int, default=1) parser.add_argument('--model_version', type=str, default='v1_13b', choices=['v1_7b', 'v1_13b', 'v2_7b', 'v2_13b']) parser.add_argument('--dtype', type=str, default='float16', choices=['float32', 'bfloat16', 'float16']) parser.add_argument('--logits_dtype', type=str, default='float32', choices=['float16', 'float32']) parser.add_argument('--output_dir', type=str, default='tllm_checkpoint', help='The path to save the TensorRT-LLM checkpoint') parser.add_argument( '--workers', type=int, default=1, help='The number of workers for converting checkpoint in parallel') parser.add_argument( '--per_channel', default=False, action="store_true", help= 'By default, we use a single static scaling factor for the GEMM\'s result. ' 'per_channel instead uses a different static scaling factor for each channel. ' 'The latter is usually more accurate, but a little slower.') parser.add_argument( '--per_token', default=False, action="store_true", help= 'By default, we use a single static scaling factor to scale activations in the int8 range. ' 'per_token chooses at run time, and for each token, a custom scaling factor. ' 'The latter is usually more accurate, but a little slower.') parser.add_argument( "--smoothquant", "-sq", type=float, default=None, help="Set the α parameter (see https://arxiv.org/pdf/2211.10438.pdf)" " to Smoothquant the model, and output int8 weights." " A good first try is 0.5. Must be in [0, 1]") parser.add_argument( '--use_weight_only', default=False, action="store_true", help='Quantize weights for the various GEMMs to INT4/INT8.' 'See --weight_only_precision to set the precision') parser.add_argument( '--weight_only_precision', const='int8', type=str, nargs='?', default='int8', choices=['int8', 'int4', 'int4_gptq'], help= 'Define the precision for the weights when using weight-only quantization.' 'You must also use --use_weight_only for that argument to have an impact.' ) parser.add_argument('--group_size', type=int, default=128, help='Group size used in GPTQ/AWQ quantization.') parser.add_argument( '--int8_kv_cache', default=False, action="store_true", help= 'By default, we use dtype for KV cache. int8_kv_cache chooses int8 quantization for KV' ) args = parser.parse_args() return args def generate_int8(weights, act_range, is_qkv=False, multi_query_mode=False): """ This function has two purposes: - compute quantized weights, scaled either per-tensor or per-column - compute scaling factors Depending on the GEMM API (CUTLASS/CUBLAS) the required scaling factors differ. CUTLASS uses two sets of scaling factors. One for the activation X, one for the weight W. CUBLAS only has one (we can't do per-row scaling). So we must provide pre-multiplied scaling factor. Here is the list of what we need (T means per-tensor, C per-column): - scale_x_orig_quant puts fp activation into the quantized range (i.e. [-128, 127], for int8). Used before the GEMM. (T) - scale_y_quant_orig puts quantized activation into the fp range. Used if the GEMM outputs int8. (T) - scale_w_quant_orig puts weights from quant range to fp range (used with CUTLASS) (T, C) - scale_y_accum_quant puts the GEMM result (XW) from accumulation range (int32) to quant range (int8) (used for CUBLAS) (T, C) Note that we don't do anything special about row-parallel GEMM. Theoretically, we could have per-GPU scaling factors too, but then the model would change depending on the number of GPUs used. For QKV projection, the behavior is special. Even if we have a single matrix to perform QKV projection, we consider it as three different matrices: Q, K, and V. So per-tensor actually means one scaling factor for each Q, K and V. For our GEMM implementation to respect this behavior, we use per-column mode and replicate values along columns. """ # compute weight scaling factors for fp->int8 and int8->fp if is_qkv and not multi_query_mode: scale_w_orig_quant_t = 127. / act_range["w"].reshape(3, -1).max( dim=-1, keepdims=True)[0].cpu().numpy() scale_w_orig_quant_c = 127. / act_range["w"].reshape(3, -1).cpu().numpy() elif is_qkv and multi_query_mode: hidden_dim = weights.shape[0] local_dim = act_range["w"].shape[0] kv_dim = (local_dim - hidden_dim) // 2 scale_w_q = act_range["w"][0:hidden_dim] scale_w_k = act_range["w"][hidden_dim:hidden_dim + kv_dim] scale_w_v = act_range["w"][-kv_dim:] scale_w_qkv_t = torch.concat([ scale_w_q.max(dim=0, keepdim=True)[0], scale_w_k.max(dim=0, keepdim=True)[0], scale_w_v.max(dim=0, keepdim=True)[0] ]) scale_w_orig_quant_t = 127. / scale_w_qkv_t.cpu().numpy() scale_w_orig_quant_c = 127. / act_range["w"].cpu().numpy() else: scale_w_orig_quant_t = 127. / act_range["w"].max().cpu().numpy() scale_w_orig_quant_c = 127. / act_range["w"].cpu().numpy() scale_w_quant_orig_t = 1.0 / scale_w_orig_quant_t scale_w_quant_orig_c = 1.0 / scale_w_orig_quant_c # compute the rest of needed scaling factors scale_x_orig_quant_t = np.array(127. / act_range["x"].max().item()) scale_y_orig_quant_t = np.array(127. / act_range["y"].max().item()) scale_y_quant_orig_t = np.array(act_range["y"].max().item() / 127.) scale_y_accum_quant_t = scale_y_orig_quant_t / (scale_x_orig_quant_t * scale_w_orig_quant_t) scale_y_accum_quant_c = scale_y_orig_quant_t / (scale_x_orig_quant_t * scale_w_orig_quant_c) if is_qkv and not multi_query_mode: scale_y_accum_quant_t = np.broadcast_to(scale_y_accum_quant_t, scale_w_orig_quant_c.shape) scale_w_quant_orig_t = np.broadcast_to(scale_w_quant_orig_t, scale_w_orig_quant_c.shape) if is_qkv and multi_query_mode: scale_q_y_accum_t = np.broadcast_to(scale_y_accum_quant_t[0], scale_w_q.shape) scale_k_y_accum_t = np.broadcast_to(scale_y_accum_quant_t[1], scale_w_k.shape) scale_v_y_accum_t = np.broadcast_to(scale_y_accum_quant_t[2], scale_w_v.shape) scale_y_accum_quant_t = np.concatenate( [scale_q_y_accum_t, scale_k_y_accum_t, scale_v_y_accum_t]) scale_w_quant_orig_t = np.concatenate([ np.broadcast_to(scale_w_quant_orig_t[0], scale_w_q.shape), np.broadcast_to(scale_w_quant_orig_t[1], scale_w_k.shape), np.broadcast_to(scale_w_quant_orig_t[2], scale_w_v.shape) ]) to_i8 = lambda x: x.round().clip(-127, 127).astype(np.int8) if is_qkv and multi_query_mode: scale_w_quant_orig_t_expand = np.ones([weights.shape[-1]]) scale_w_quant_orig_t_expand[:hidden_dim] = scale_w_quant_orig_t[0] scale_w_quant_orig_t_expand[hidden_dim:hidden_dim + kv_dim] = scale_w_quant_orig_t[1] scale_w_quant_orig_t_expand[-kv_dim:] = scale_w_quant_orig_t[2] weight_int8 = to_i8(weights * scale_w_quant_orig_t_expand) else: weight_int8 = to_i8(weights * scale_w_orig_quant_t) return { "weight.int8": weight_int8, "weight.int8.col": to_i8(weights * scale_w_orig_quant_c), "scale_x_orig_quant": scale_x_orig_quant_t.astype(np.float32), "scale_w_quant_orig": scale_w_quant_orig_t.astype(np.float32), "scale_w_quant_orig.col": scale_w_quant_orig_c.astype(np.float32), "scale_y_accum_quant": scale_y_accum_quant_t.astype(np.float32), "scale_y_accum_quant.col": scale_y_accum_quant_c.astype(np.float32), "scale_y_quant_orig": scale_y_quant_orig_t.astype(np.float32), } @torch.no_grad() def apply_smoothing(scales, gemm_weights, layernorm_weights=None, layernorm_bias=None, dtype=torch.float32, layernorm_1p=False): if not isinstance(gemm_weights, list): gemm_weights = [gemm_weights] if layernorm_weights is not None: assert layernorm_weights.numel() == scales.numel() layernorm_weights.div_(scales).to(dtype) if layernorm_bias is not None: assert layernorm_bias.numel() == scales.numel() layernorm_bias.div_(scales).to(dtype) if layernorm_1p: layernorm_weights += (1 / scales) - 1 for gemm in gemm_weights: gemm.mul_(scales.view(1, -1)).to(dtype) @torch.no_grad() def smooth_gemm(gemm_weights, act_scales, layernorm_weights=None, layernorm_bias=None, alpha=0.5, weight_scales=None): if not isinstance(gemm_weights, list): gemm_weights = [gemm_weights] orig_dtype = gemm_weights[0].dtype for gemm in gemm_weights: # gemm_weights are expected to be transposed assert gemm.shape[1] == act_scales.numel() if weight_scales is None: weight_scales = torch.cat( [gemm.abs().max(dim=0, keepdim=True)[0] for gemm in gemm_weights], dim=0) weight_scales = weight_scales.max(dim=0)[0] weight_scales.to(float).clamp(min=1e-5) scales = (act_scales.to(gemm_weights[0].device).to(float).pow(alpha) / weight_scales.pow(1 - alpha)).clamp(min=1e-5) apply_smoothing(scales, gemm_weights, layernorm_weights, layernorm_bias, orig_dtype) return scales @torch.no_grad() def smooth_gemm_fc1_gate(fc1_weights, gate_weights, act_scales, layernorm_weights=None, layernorm_bias=None, alpha=0.5, weight_scales=None): gemm_weights = [] if not isinstance(fc1_weights, list): fc1_weights = [fc1_weights] if not isinstance(gate_weights, list): gate_weights = [gate_weights] for i in range(len(fc1_weights)): gemm_weight = torch.cat([fc1_weights[i], gate_weights[i]], dim=0) gemm_weights.append(gemm_weight) orig_dtype = gemm_weights[0].dtype for gemm in gemm_weights: # gemm_weights are expected to be transposed assert gemm.shape[1] == act_scales.numel() if weight_scales is None: weight_scales = torch.cat( [gemm.abs().max(dim=0, keepdim=True)[0] for gemm in gemm_weights], dim=0) weight_scales = weight_scales.max(dim=0)[0] weight_scales.to(float).clamp(min=1e-5) scales = (act_scales.to(gemm_weights[0].device).to(float).pow(alpha) / weight_scales.pow(1 - alpha)).clamp(min=1e-5) apply_smoothing(scales, fc1_weights + gate_weights, layernorm_weights, layernorm_bias, orig_dtype) return scales @torch.no_grad() def capture_activation_range(model, tokenizer, num_samples=512, seq_len=512): model.eval() next(model.parameters()).device act_scales = defaultdict(lambda: {"x": None, "y": None, "w": None}) test_token_num = 923 tokenizer.pad_token = tokenizer.eos_token def stat_tensor(name, tensor, act_scales, key): hidden_dim = tensor.shape[-1] tensor = tensor.view(-1, hidden_dim).abs().detach() comming_max = torch.max(tensor, dim=0)[0].float() if act_scales[name][key] is None: act_scales[name][key] = comming_max else: act_scales[name][key] = torch.max(act_scales[name][key], comming_max) def stat_input_hook(m, x, y, name): if isinstance(x, tuple): x = x[0] stat_tensor(name, x, act_scales, "x") stat_tensor(name, y, act_scales, "y") if act_scales[name]["w"] is None: act_scales[name]["w"] = m.weight.abs().clip(1e-8, None).max(dim=1)[0] hooks = [] for name, m in model.named_modules(): if isinstance(m, nn.Linear) or isinstance(m, Conv1D): hooks.append( m.register_forward_hook( functools.partial(stat_input_hook, name=name))) from datasets import load_dataset dataset_cnn = load_dataset("ccdv/cnn_dailymail", '3.0.0') for i in tqdm(range(num_samples), desc="calibrating model"): datapoint = dataset_cnn['train'][i:i + 1] line = copy.copy(datapoint['article']) line[0] = line[0] + ' TL;DR: ' line[0] = line[0].strip() line[0] = line[0].replace(" n't", "n't") line_encoded = tokenizer(line, return_tensors="pt", padding=True, truncation=True)["input_ids"].type(torch.int64) line_encoded = line_encoded[:, -test_token_num:] line_encoded = line_encoded.cuda() model(line_encoded) for h in hooks: h.remove() return act_scales @torch.no_grad() def smooth_baichuan_model(model, scales, alpha, baichuan_smoother): # Smooth the activation and weights with smoother = $\diag{s}$ for name, module in model.named_modules(): class_name = module.__class__.__name__ if not 'Layer' in class_name: continue print(f'smoothing module: {name}, class_name: {class_name}') # qkv_proj layer_name_qkv = name + ".self_attn.W_pack" smoother = smooth_gemm(module.self_attn.W_pack.weight, scales[layer_name_qkv]["x"], module.input_layernorm.weight, None, alpha) scales[layer_name_qkv]["x"] = scales[layer_name_qkv]["x"] / smoother scales[layer_name_qkv]["w"] = module.self_attn.W_pack.weight.abs().max( dim=1)[0].float() # ================================================================= layer_name = name + ".self_attn.o_proj" smoother = smooth_gemm(module.self_attn.o_proj.weight, scales[layer_name]["x"], None, None, alpha) baichuan_smoother[layer_name] = smoother.float() scales[layer_name]["x"] = scales[layer_name]["x"] / smoother scales[layer_name]["w"] = module.self_attn.o_proj.weight.abs().max( dim=1)[0].float() # ================================================================== fc1_layer_name = name + ".mlp.gate_proj" gate_layer_name = name + ".mlp.up_proj" smoother = smooth_gemm_fc1_gate(module.mlp.gate_proj.weight, module.mlp.up_proj.weight, scales[fc1_layer_name]["x"], module.post_attention_layernorm.weight, None, alpha) scales[fc1_layer_name]["x"] = scales[fc1_layer_name]["x"] / smoother scales[fc1_layer_name]["w"] = module.mlp.gate_proj.weight.abs().max( dim=1)[0].float() scales[gate_layer_name]["x"] = scales[gate_layer_name]["x"] / smoother scales[gate_layer_name]["w"] = module.mlp.up_proj.weight.abs().max( dim=1)[0].float() # ================================================================== layer_name = name + ".mlp.down_proj" smoother = smooth_gemm(module.mlp.down_proj.weight, scales[layer_name]["x"], None, None, alpha) baichuan_smoother[layer_name] = smoother.float() scales[layer_name]["x"] = scales[layer_name]["x"] / smoother scales[layer_name]["w"] = module.mlp.down_proj.weight.abs().max( dim=1)[0].float() def get_tllm_linear_sq_weight(vals, prefix, shape, tensor_parallel, is_qkv=False, per_token=False, per_channel=False, last_prefix=None, bias=None, smoother_value=None, smoother_shape=None, rank=0, cat_dim=0, multi_query_mode=False): results = {} def multi_query_split(data, local_dim, head_size, tp_size, cur_rank): q, k, v = np.split(data, [local_dim, local_dim + head_size], axis=-1) q_split = np.split(q, tp_size, axis=-1) k_split = np.split(k, tp_size, axis=-1) v_split = np.split(v, tp_size, axis=-1) return [ np.concatenate((q_split[ii], k_split[ii], v_split[ii]), axis=-1) for ii in range(tp_size) ][cur_rank] col_shape = shape if (is_qkv or per_channel) else [1, 1] if per_token: if per_channel: original_weights = np.array(vals["weight.int8.col"]) else: original_weights = np.array(vals["weight.int8"]) local_dim = original_weights.shape[0] head_size = (original_weights.shape[1] - local_dim) // 2 if multi_query_mode: cur_weights = multi_query_split(original_weights, local_dim, head_size, tensor_parallel, rank) else: cur_weights = np.split(original_weights, tensor_parallel, axis=cat_dim)[rank] if is_qkv: hidden_dim = cur_weights.shape[0] cur_weights = cur_weights.reshape(hidden_dim, -1) results[prefix + 'weight'] = torch.from_numpy(cur_weights).t().contiguous() if smoother_value is None: results[last_prefix] = torch.from_numpy( np.array([1.0], dtype=np.float32)) if per_channel: cur_per_channel_value = vals["scale_w_quant_orig.col"] if smoother_value is None: if multi_query_mode: cur_per_channel_value = multi_query_split( vals["scale_w_quant_orig.col"], local_dim, head_size, tensor_parallel, rank) else: cur_per_channel_value = np.split( vals["scale_w_quant_orig.col"], tensor_parallel, axis=cat_dim)[rank] else: cur_per_channel_value = vals["scale_w_quant_orig"] if is_qkv: if multi_query_mode: cur_per_channel_value = multi_query_split( vals["scale_w_quant_orig"], local_dim, head_size, tensor_parallel, rank) else: cur_per_channel_value = np.split(vals["scale_w_quant_orig"], tensor_parallel, axis=cat_dim)[rank] results[prefix + 'per_channel_scale'] = torch.from_numpy( np.array(cur_per_channel_value, dtype=np.float32).reshape(col_shape)).contiguous() else: if per_channel: original_weights = np.array(vals["weight.int8.col"]) else: original_weights = np.array(vals["weight.int8"]) local_dim = original_weights.shape[0] head_size = (original_weights.shape[1] - local_dim) // 2 if multi_query_mode: cur_weights = multi_query_split(original_weights, local_dim, head_size, tensor_parallel, rank) else: cur_weights = np.split(original_weights, tensor_parallel, axis=cat_dim)[rank] if is_qkv: hidden_dim = cur_weights.shape[0] cur_weights = cur_weights.reshape(hidden_dim, -1) results[prefix + 'weight'] = torch.from_numpy(cur_weights).t().contiguous() if per_channel: cur_per_channel_value = vals["scale_y_accum_quant.col"] if smoother_value is None: if multi_query_mode: cur_per_channel_value = multi_query_split( vals["scale_y_accum_quant.col"], local_dim, head_size, tensor_parallel, rank) else: cur_per_channel_value = np.split( vals["scale_y_accum_quant.col"], tensor_parallel, axis=cat_dim)[rank] else: cur_per_channel_value = vals["scale_y_accum_quant"] # QKV is always per_channel if is_qkv: if multi_query_mode: cur_per_channel_value = multi_query_split( vals["scale_y_accum_quant"], local_dim, head_size, tensor_parallel, rank) else: cur_per_channel_value = np.split( vals["scale_y_accum_quant"], tensor_parallel, axis=cat_dim)[rank] results[prefix + 'per_channel_scale'] = torch.from_numpy( np.array([cur_per_channel_value], dtype=np.float32).reshape(col_shape)).contiguous() results[last_prefix] = torch.from_numpy( np.array([vals['scale_x_orig_quant']], dtype=np.float32)).contiguous() results[prefix + 'act_scale'] = torch.from_numpy( np.array([[vals["scale_y_quant_orig"]]], dtype=np.float32)).contiguous() if smoother_value is not None: cur_smoother_value = np.split(smoother_value, tensor_parallel, axis=cat_dim)[rank] results[prefix + 'smoother'] = cur_smoother_value.reshape( smoother_shape).contiguous().to(torch.float32) if bias is not None: results[prefix + 'bias'] = bias return results def split(weight: torch.Tensor, tp_size: int, rank: int = 0, dim: int = 0) -> torch.Tensor: if tp_size == 1: return weight elif weight.ndim == 1: return torch.chunk(weight, tp_size)[rank].contiguous() else: return torch.chunk(weight, tp_size, dim=dim)[rank].contiguous() def split_qkv_tp(qkv, n_head, n_kv_heads, n_hidden, tensor_parallel, rank): """ Splits the QKV matrix according to tensor parallelism """ kv_head_size = n_kv_heads * (n_hidden // n_head) q, k, v = torch.split(qkv, [n_hidden, kv_head_size, kv_head_size], dim=0) q = split(q, tensor_parallel, rank, dim=0) k = split(k, tensor_parallel, rank, dim=0) v = split(v, tensor_parallel, rank, dim=0) return torch.concatenate([q, k, v], dim=0).contiguous() def split_matrix(weight: torch.Tensor, tp_size: int, rank: int, dim: int) -> torch.Tensor: return split(weight, tp_size, rank, dim=dim) def get_weight(params: Dict[str, torch.Tensor], prefix: str, dtype: torch.dtype) -> torch.Tensor: if f'{prefix}.weight' not in params: return None return params[f'{prefix}.weight'].to(dtype).detach().cpu() def get_bias(params: Dict[str, torch.Tensor], prefix: str, dtype: torch.dtype) -> torch.Tensor: if f'{prefix}.bias' not in params: return None return params[f'{prefix}.bias'].to(dtype).detach().cpu() def get_weight_and_bias(params: Dict[str, torch.Tensor], prefix: str, dtype: torch.dtype) -> Tuple[torch.Tensor]: return get_weight(params, prefix, dtype), get_bias(params, prefix, dtype) def get_tllm_linear_weight( weight: torch.Tensor, prefix: str, bias: Optional[torch.Tensor] = None, ) -> Dict[str, torch.Tensor]: results = {} results[f'{prefix}.weight'] = weight.contiguous() if bias is not None: results[f'{prefix}.bias'] = bias return results def load_baichuan_config(model_dir: str) -> AutoConfig: config = AutoConfig.from_pretrained(model_dir, trust_remote_code=True) return config def convert_hf_baichuan_sq(hf_model, mapping, rank=0, dtype='float32', per_channel=False, per_token=False, int8_kv_cache=False, act_range=[], smoother=[]): weights = {} tik = time.time() tensor_parallel = mapping.tp_size model_params = dict(hf_model.named_parameters()) dtype = getattr(torch, dtype) num_attention_heads = hf_model.config.num_attention_heads hidden_size = hf_model.config.hidden_size inter_size = hf_model.config.intermediate_size num_key_value_heads = hf_model.config.num_attention_heads multi_query_mode = (num_key_value_heads != num_attention_heads) for l in range(hf_model.config.num_hidden_layers): prefix = f'model.layers.{l}.' tllm_prex = f'transformer.layers.{l}.' # self_attn.W_pack -> attention.qkv qkv_weight = get_weight(model_params, prefix + 'self_attn.W_pack', dtype) qkv_weight = qkv_weight.t().numpy() qkv_out_dim = qkv_weight.shape[1] if not multi_query_mode: qkv_weight = qkv_weight.reshape(hidden_size, 3, hidden_size) int8_weights = generate_int8(qkv_weight, act_range.get(prefix + 'self_attn.W_pack'), is_qkv=True, multi_query_mode=multi_query_mode) weights.update( get_tllm_linear_sq_weight(int8_weights, tllm_prex + 'attention.qkv.', [1, qkv_out_dim // tensor_parallel], tensor_parallel, is_qkv=True, per_token=per_token, per_channel=per_channel, last_prefix=tllm_prex + 'input_layernorm.scale_to_int', smoother_value=None, smoother_shape=None, rank=rank, cat_dim=-1, multi_query_mode=multi_query_mode)) if int8_kv_cache: qkv_weight = get_weight(model_params, prefix + 'self_attn.W_pack', dtype) qkv_weight = qkv_weight.t().numpy() if not multi_query_mode: qkv_weight = qkv_weight.reshape(hidden_size, 3, hidden_size) int8_weights = generate_int8(qkv_weight, act_range.get(prefix + 'self_attn.W_pack'), is_qkv=True, multi_query_mode=multi_query_mode) weights[tllm_prex + 'attention.kv_cache_scaling_factor'] = torch.from_numpy( np.array([int8_weights['scale_y_quant_orig']], dtype=np.float32)).contiguous() # attn.out_proj -> attention.dense attn_dense_weight = get_weight(model_params, prefix + 'self_attn.o_proj', dtype) attn_dense_weight = attn_dense_weight.t().numpy() int8_weights = generate_int8(attn_dense_weight, act_range.get(prefix + 'self_attn.o_proj')) weights.update( get_tllm_linear_sq_weight( int8_weights, tllm_prex + 'attention.dense.', [1, hidden_size], tensor_parallel, is_qkv=False, per_token=per_token, per_channel=per_channel, last_prefix=tllm_prex + 'attention.quantization_scaling_factor', smoother_value=smoother[(prefix + 'self_attn.o_proj')], smoother_shape=[1, hidden_size // tensor_parallel], rank=rank, cat_dim=0)) # mlp.gate_proj -> mlp.fc mlp_fc_weight = get_weight(model_params, prefix + 'mlp.gate_proj', dtype) mlp_fc_weight = mlp_fc_weight.t().numpy() int8_weights = generate_int8(mlp_fc_weight, act_range.get(prefix + 'mlp.gate_proj')) weights.update( get_tllm_linear_sq_weight( int8_weights, tllm_prex + 'mlp.fc.', [1, inter_size // tensor_parallel], tensor_parallel, is_qkv=False, per_token=per_token, per_channel=per_channel, last_prefix=tllm_prex + 'post_layernorm.scale_to_int', smoother_value=None, smoother_shape=None, rank=rank, cat_dim=-1)) # mlp.down_proj -> mlp.proj mlp_proj_weight = get_weight(model_params, prefix + 'mlp.down_proj', dtype) mlp_proj_weight = mlp_proj_weight.t().numpy() int8_weights = generate_int8(mlp_proj_weight, act_range.get(prefix + 'mlp.down_proj')) weights.update( get_tllm_linear_sq_weight( int8_weights, tllm_prex + 'mlp.proj.', [1, hidden_size], tensor_parallel, is_qkv=False, per_token=per_token, per_channel=per_channel, last_prefix=tllm_prex + 'mlp.quantization_scaling_factor', smoother_value=smoother[prefix + 'mlp.down_proj'], smoother_shape=[1, inter_size // tensor_parallel], rank=rank, cat_dim=0)) # mlp.up_proj -> mlp.gate mlp_gate_weight = get_weight(model_params, prefix + 'mlp.up_proj', dtype) mlp_gate_weight = mlp_gate_weight.t().numpy() int8_weights = generate_int8(mlp_gate_weight, act_range.get(prefix + 'mlp.up_proj')) weights.update( get_tllm_linear_sq_weight( int8_weights, tllm_prex + 'mlp.gate.', [1, inter_size // tensor_parallel], tensor_parallel, is_qkv=False, per_token=per_token, per_channel=per_channel, last_prefix=tllm_prex + 'post_layernorm.scale_to_int', smoother_value=None, smoother_shape=None, rank=rank, cat_dim=-1)) # input layer_norm input_ln_weight = get_weight(model_params, prefix + 'input_layernorm', dtype) weights[tllm_prex + 'input_layernorm.weight'] = input_ln_weight # post layer_norm post_ln_weight = get_weight(model_params, prefix + 'post_attention_layernorm', dtype) weights[tllm_prex + 'post_layernorm.weight'] = post_ln_weight embed_w = get_weight(model_params, 'model.embed_tokens', dtype) if mapping.is_first_pp_rank(): # Embedding weights['transformer.vocab_embedding.weight'] = embed_w lm_head_w = get_weight(model_params, 'lm_head', dtype) if mapping.is_last_pp_rank(): # lm_head weight and bias weights['lm_head.weight'] = split_matrix(lm_head_w.clone(), mapping.tp_size, mapping.tp_rank, dim=0) ln_f_w = get_weight(model_params, 'model.norm', dtype) # ln_f weight and bias weights['transformer.ln_f.weight'] = ln_f_w tok = time.time() t = time.strftime('%H:%M:%S', time.gmtime(tok - tik)) print(f'Weights loaded. Total time: {t}') return weights def convert_hf_baichuan(hf_model: AutoModelForCausalLM, hf_config: AutoConfig, model_version: str, mapping: Mapping, dtype: str = 'float32', quant_algo: str = None): weights = {} tik = time.time() model_params = dict(hf_model.named_parameters()) dtype = getattr(torch, dtype) num_hidden_layers = hf_config.num_hidden_layers hf_key = [ "model.embed_tokens.weight", # vocab_embedding "lm_head.weight", # lm_head "model.norm.weight", # ln_f "self_attn.W_pack.weight", # attention.qkv "self_attn.o_proj.weight", # attention.dense "mlp.up_proj.weight", # mlp.gate "mlp.down_proj.weight", # mlp.proj "mlp.gate_proj.weight", # mlp.fc "input_layernorm.weight", # input_layernorm "post_attention_layernorm.weight", # post_layernorm ] def load(key_id, layer_idx=-1, tp_dim=-1, quant=False): prefix = "" if layer_idx == -1 else f"model.layers.{layer_idx}." v = model_params[prefix + hf_key[key_id]] if key_id == 3: q_emb = v.shape[0] // 3 model_emb = v.shape[1] v = v.reshape(3, q_emb, model_emb) if v.shape[1] % mapping.tp_size != 0: tensorrt_llm.logger.error( "Current weight shape is invalid for mapping.tp_size=" + str(mapping.tp_size)) v = v.split(v.shape[1] // mapping.tp_size, dim=1)[mapping.tp_rank] v = v.reshape(3 * (q_emb // mapping.tp_size), model_emb) if tp_dim >= 0: if v.shape[tp_dim] % mapping.tp_size != 0: tensorrt_llm.logger.error( "Current weight shape is invalid for mapping.tp_size=" + str(mapping.tp_size)) v = v.split(v.shape[tp_dim] // mapping.tp_size, dim=tp_dim)[mapping.tp_rank] v = v.to(dtype).contiguous().detach().cpu() return v # Convert vocab_embedding if mapping.is_first_pp_rank(): weights['transformer.vocab_embedding.weight'] = load(0) # Convert lm_head v = load(1, -1, 0) if model_version.startswith('v2'): v = torch.nn.functional.normalize(v) if mapping.is_last_pp_rank(): weights['lm_head.weight'] = v # Convert ln_f if mapping.is_last_pp_rank(): weights['transformer.ln_f.weight'] = load(2) # Convert layers layers_range = mapping.pp_layers(num_hidden_layers) for l in layers_range: prefix = f"transformer.layers.{l}." weights[prefix + 'attention.qkv.weight'] = load(3, l) weights[prefix + 'attention.dense.weight'] = load(4, l, 1) weights[prefix + 'mlp.gate.weight'] = load(5, l, 0) weights[prefix + 'mlp.proj.weight'] = load(6, l, 1) weights[prefix + 'mlp.fc.weight'] = load(7, l, 0) weights[prefix + 'input_layernorm.weight'] = load(8, l) weights[prefix + 'post_layernorm.weight'] = load(9, l) tok = time.time() t = time.strftime('%H:%M:%S', time.gmtime(tok - tik)) print(f'Weights loaded. Total time: {t}') return weight_only_quantize_dict(weights, quant_algo=quant_algo, plugin=True) def convert_baichuan_gptq(hf_config: AutoConfig, quant_ckpt_path: str, model_version: str, mapping=Mapping(), dtype="float16"): tensorrt_llm.logger.info( 'Loading weights from groupwise GPTQ Baichuan safetensors...') weights = {} tik = time.time() gptq_baichuan = safetensors.safe_open(quant_ckpt_path, framework="pt", device=0) gptq_prefix = "model." gptq_suffix_list = [".qweight", ".qzeros", ".scales"] gptq_key_list = [ "embed_tokens.weight", # vocab_embedding "lm_head.weight", # lm_head "norm.weight", # ln_f "self_attn.W_pack", # attention.qkv "_proj", # "self_attn.o_proj", # attention.dense "mlp.up_proj", # mlp.gate "mlp.down_proj", # mlp.proj "mlp.gate_proj", # mlp.fc "input_layernorm.weight", # input_layernorm "post_attention_layernorm.weight", # post_layernorm ] packer = torch.ops.trtllm.pack_int8_tensor_to_packed_int4 preprocessor = torch.ops.trtllm.preprocess_weights_for_mixed_gemm torch_dtype = getattr(torch, dtype) def load(key, no_prefix=0): if no_prefix: return gptq_baichuan.get_tensor(key) else: return gptq_baichuan.get_tensor(gptq_prefix + key) def torch_split(v, dim): if v.shape[dim] % mapping.tp_size != 0: tensorrt_llm.logger.error( "Current weight shape is invalid for mapping.tp_size=" + str(mapping.tp_size)) assert False, "Invalid TP size" return v.split(v.shape[dim] // mapping.tp_size, dim=dim)[mapping.tp_rank] def unpack_int32_into_int8(w_packed): # Unpack inputs packed in int32/float32 into uint4 and store them in int8 format w_packed_int4x2 = w_packed.contiguous().view(torch.uint8) w_unpacked = torch.zeros(w_packed_int4x2.shape[0], w_packed_int4x2.shape[1] * 2, dtype=torch.int8) w_unpacked[:, ::2] = w_packed_int4x2 % 16 w_unpacked[:, 1::2] = w_packed_int4x2 // 16 return w_unpacked.contiguous() def process_and_assign_weight(prefix, v, tp_dim=-1): if tp_dim == -1: qweight_int32, qzeros_int32, scales_fp16 = [ item.cpu() for item in v ] else: qweight_int32, qzeros_int32, scales_fp16 = [ torch_split(item, tp_dim).cpu() for item in v ] USE_UINT4_INPUT = 1 # Set to true if checkpoint store UINT4 weights USE_GPTQ_FOR_LLAMA = 1 # GPTQ-for-LLaMA added 1 to zeros qweight_unpacked_int8 = unpack_int32_into_int8( qweight_int32.T).T.contiguous() - 8 qweight_interleaved = preprocessor(packer(qweight_unpacked_int8), torch.quint4x2, torch.float16).view(torch.float16) # zeros = zeros * scales qzeros_unpacked_int32 = unpack_int32_into_int8(qzeros_int32) if not USE_UINT4_INPUT: # Correcting UINT4 values back to INT4 order mask_negative = qzeros_unpacked_int32[qzeros_unpacked_int32 < 0] mask_positive = qzeros_unpacked_int32[qzeros_unpacked_int32 >= 0] qzeros_unpacked_int32 = qzeros_unpacked_int32 + 16 * mask_negative - 16 * mask_positive zeros_x_scales_fp16 = (-qzeros_unpacked_int32 + 8 * USE_UINT4_INPUT - USE_GPTQ_FOR_LLAMA) * scales_fp16 zeros_x_scales_fp16 = zeros_x_scales_fp16.half() # return processed interleaved weight, original scales and zeros * scales weights[prefix + ".weight"] = qweight_interleaved weights[prefix + ".weights_scaling_factor"] = scales_fp16 weights[prefix + ".zero"] = zeros_x_scales_fp16 # Load weights from GPTQ checkpoint into TRT-LLM module # 1. vocab_embedding v = load(gptq_key_list[0]) if mapping.is_first_pp_rank(): weights['transformer.vocab_embedding.weight'] = v.to(torch_dtype) # 2. lm_head original_v = load(gptq_key_list[1], "no_prefix") if model_version.startswith('v2'): # baichuan v2 models use NormHead tensorrt_llm.logger.info( f'Normalizing lm_head.weight for {model_version}') v = torch_split(torch.nn.functional.normalize(original_v), 0) else: v = torch_split(original_v, 0) if mapping.is_last_pp_rank(): weights['lm_head.weight'] = v.to(torch_dtype) # 3. ln_f v = load(gptq_key_list[2]) if mapping.is_last_pp_rank(): weights['transformer.ln_f.weight'] = v.to(torch_dtype) # 4. Weights inside each layer num_hidden_layers = hf_config.num_hidden_layers layers_range = mapping.pp_layers(num_hidden_layers) for l in layers_range: layer_idx = l - layers_range[0] prefix = f"layers.{l}." tllm_prefix = f"transformer.layers.{l}." tensorrt_llm.logger.info(f'Process weights in layer: {layer_idx}') # 4.1 attention.qkv qkv_weight_list = [] for suf in gptq_suffix_list: qkv_list = [] comp_part = load(prefix + gptq_key_list[3] + suf) qkv = torch.chunk(comp_part, 3, 1) for i in range(3): comp_part = qkv[i] comp_part = torch_split(comp_part, 1) qkv_list.append(comp_part) qkv_weight_list.append(torch.cat(qkv_list, dim=1)) process_and_assign_weight(tllm_prefix + "attention.qkv", qkv_weight_list) # 4.2 attention.dense v = [load(prefix + gptq_key_list[5] + suf) for suf in gptq_suffix_list] process_and_assign_weight(tllm_prefix + "attention.dense", v, 0) # 4.3 mlp.gate v = [load(prefix + gptq_key_list[6] + suf) for suf in gptq_suffix_list] process_and_assign_weight(tllm_prefix + "mlp.gate", v, 1) # 4.4 mlp.proj v = [load(prefix + gptq_key_list[7] + suf) for suf in gptq_suffix_list] process_and_assign_weight(tllm_prefix + "mlp.proj", v, 0) # 4.5 mlp.fc v = [load(prefix + gptq_key_list[8] + suf) for suf in gptq_suffix_list] process_and_assign_weight(tllm_prefix + "mlp.fc", v, 1) # 4.6 input_layernorm v = load(prefix + gptq_key_list[9]) weights[tllm_prefix + 'input_layernorm.weight'] = v.to(torch_dtype) # 4.7 pst_layernorm v = load(prefix + gptq_key_list[10]) weights[tllm_prefix + 'post_layernorm.weight'] = v.to(torch_dtype) tok = time.time() t = time.strftime('%H:%M:%S', time.gmtime(tok - tik)) tensorrt_llm.logger.info(f'Weights loaded. Total time: {t}') return weights if __name__ == '__main__': # TODO(qijun): Currently, the convert script depends on a torch op: # torch.ops.fastertransformer.symmetric_quantize_last_axis_of_batched_matrix, # which is included in tensorrt_llm Python package. Otherwise, the convert # script does not need to import tensorrt_llm. Will remove it after reimplementing # the op with PyTorch. print(tensorrt_llm.__version__) args = parse_arguments() tik = time.time() if not os.path.exists(args.output_dir): os.makedirs(args.output_dir) quant_algo = None plugin_weight_only_quant_type = None if args.use_weight_only and args.weight_only_precision == 'int8': plugin_weight_only_quant_type = torch.int8 quant_algo = QuantAlgo.W8A16 elif args.use_weight_only and args.weight_only_precision == 'int4': plugin_weight_only_quant_type = torch.quint4x2 quant_algo = QuantAlgo.W4A16 elif args.use_weight_only and args.weight_only_precision == 'int4_gptq': quant_algo = QuantAlgo.W4A16_GPTQ if args.smoothquant: if args.per_token and args.per_channel: quant_algo = QuantAlgo.W8A8_SQ_PER_CHANNEL_PER_TOKEN_PLUGIN elif not args.per_token and not args.per_channel: quant_algo = QuantAlgo.W8A8_SQ_PER_TENSOR_PLUGIN elif not args.per_token and args.per_channel: quant_algo = QuantAlgo.W8A8_SQ_PER_CHANNEL_PER_TENSOR_PLUGIN elif args.per_token and not args.per_channel: quant_algo = QuantAlgo.W8A8_SQ_PER_TENSOR_PER_TOKEN_PLUGIN if args.int8_kv_cache: kv_cache_quant_algo = QuantAlgo.INT8 else: kv_cache_quant_algo = None if args.model_version == 'v1_7b' or args.model_version == 'v2_7b': position_embedding_type = 'rope_gpt_neox' else: position_embedding_type = 'alibi' hf_config = load_baichuan_config(args.model_dir) if args.model_version == 'v1_7b' or args.model_version == 'v2_7b': max_position_embeddings = hf_config.max_position_embeddings else: max_position_embeddings = hf_config.model_max_length if args.weight_only_precision == 'int4_gptq': hf_config.vocab_size = int((hf_config.vocab_size + 63) / 64) * 64 world_size = args.tp_size * args.pp_size config = { 'architecture': 'BaichuanForCausalLM', 'dtype': args.dtype, 'logits_dtype': args.logits_dtype, 'vocab_size': hf_config.vocab_size, 'max_position_embeddings': max_position_embeddings, 'hidden_size': hf_config.hidden_size, 'num_hidden_layers': hf_config.num_hidden_layers, 'num_attention_heads': hf_config.num_attention_heads, 'num_key_value_heads': hf_config.num_attention_heads, 'hidden_act': hf_config.hidden_act, 'intermediate_size': hf_config.intermediate_size, 'norm_epsilon': hf_config.rms_norm_eps, 'position_embedding_type': position_embedding_type, 'quantization': { 'quant_algo': quant_algo, 'kv_cache_quant_algo': kv_cache_quant_algo, 'group_size': args.group_size, }, 'mapping': { 'world_size': world_size, 'tp_size': args.tp_size, 'pp_size': args.pp_size, }, } if args.use_weight_only and args.weight_only_precision == 'int4_gptq': config['quantization'].update({ 'has_zero_point': True, }) with open(os.path.join(args.output_dir, 'config.json'), 'w') as f: json.dump(config, f, indent=4) hf_model = AutoModelForCausalLM.from_pretrained(args.model_dir, trust_remote_code=True, torch_dtype="auto") if args.smoothquant is not None or args.int8_kv_cache: act_range = {} baichuan_smoother = {} act_range = capture_activation_range( hf_model.cuda(), AutoTokenizer.from_pretrained(args.model_dir, use_fast=False, trust_remote_code=True)) if args.smoothquant is not None: smooth_baichuan_model(hf_model, act_range, args.smoothquant, baichuan_smoother) def covert_and_save(rank): mapping = Mapping(world_size=world_size, rank=rank, tp_size=args.tp_size, pp_size=args.pp_size) if args.smoothquant is not None or args.int8_kv_cache: weights = convert_hf_baichuan_sq(hf_model, mapping, rank, args.dtype, args.per_channel, args.per_token, args.int8_kv_cache, act_range, baichuan_smoother) elif args.use_weight_only and args.weight_only_precision == 'int4_gptq': weights = convert_baichuan_gptq(hf_config, args.quant_ckpt_path, args.model_version, mapping, dtype=args.dtype) else: weights = convert_hf_baichuan(hf_model, hf_config, args.model_version, mapping, dtype=args.dtype, quant_algo=quant_algo) safetensors.torch.save_file( weights, os.path.join(args.output_dir, f'rank{rank}.safetensors')) if args.workers == 1: for rank in range(world_size): covert_and_save(rank) else: with ThreadPoolExecutor(max_workers=args.workers) as p: futures = [ p.submit(covert_and_save, rank) for rank in range(world_size) ] exceptions = [] for future in as_completed(futures): try: future.result() except Exception as e: traceback.print_exc() exceptions.append(e) assert len( exceptions ) == 0, "Checkpoint conversion failed, please check error log." del hf_model tok = time.time() t = time.strftime('%H:%M:%S', time.gmtime(tok - tik)) print(f'Total time of converting checkpoints: {t}')