# SPDX-FileCopyrightText: Copyright (c) 2022-2023 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 math import os import time # isort: off import torch import torch.multiprocessing as mp import tvm.tensorrt as trt # isort: on from transformers import AutoConfig, AutoModelForCausalLM try: from transformers import Qwen2ForCausalLM except ImportError: print( "Qwen1.5 requires transformers>=4.37.1, type pip install transformers==4.37.1" ) import xtrt_llm from xtrt_llm._utils import str_dtype_to_xtrt from xtrt_llm.builder import Builder from xtrt_llm.logger import logger from xtrt_llm.mapping import Mapping from xtrt_llm.models import quantize_model from xtrt_llm.network import net_guard from xtrt_llm.plugin.plugin import ContextFMHAType from xtrt_llm.quantization import QuantMode MODEL_NAME = "qwen" import onnx import tvm.tensorrt as trt from onnx import TensorProto, helper now_dir = os.path.dirname(os.path.abspath(__file__)) def trt_dtype_to_onnx(dtype): if dtype == trt.float16: return TensorProto.DataType.FLOAT16 elif dtype == trt.float32: return TensorProto.DataType.FLOAT elif dtype == trt.int32: return TensorProto.DataType.INT32 else: raise TypeError("%s is not supported" % dtype) def to_onnx(network, path): inputs = [] for i in range(network.num_inputs): network_input = network.get_input(i) inputs.append( helper.make_tensor_value_info( network_input.name, trt_dtype_to_onnx(network_input.dtype), list(network_input.shape))) outputs = [] for i in range(network.num_outputs): network_output = network.get_output(i) outputs.append( helper.make_tensor_value_info( network_output.name, trt_dtype_to_onnx(network_output.dtype), list(network_output.shape))) nodes = [] for i in range(network.num_layers): layer = network.get_layer(i) layer_inputs = [] for j in range(layer.num_inputs): ipt = layer.get_input(j) if ipt is not None: layer_inputs.append(layer.get_input(j).name) layer_outputs = [ layer.get_output(j).name for j in range(layer.num_outputs) ] nodes.append( helper.make_node(str(layer.type), name=layer.name, inputs=layer_inputs, outputs=layer_outputs, domain="com.nvidia")) onnx_model = helper.make_model(helper.make_graph(nodes, 'attention', inputs, outputs, initializer=None), producer_name='NVIDIA') onnx.save(onnx_model, path) def get_engine_name(model, dtype, tp_size, pp_size, rank): if pp_size == 1: return '{}_{}_tp{}_rank{}.engine'.format(model, dtype, tp_size, rank) return '{}_{}_tp{}_pp{}_rank{}.engine'.format(model, dtype, tp_size, pp_size, rank) def serialize_engine(engine, path): logger.info(f'Serializing engine to {path}...') tik = time.time() engine.serialize(path) tok = time.time() t = time.strftime('%H:%M:%S', time.gmtime(tok - tik)) logger.info(f'Engine serialized. Total time: {t}') def parse_arguments(): parser = argparse.ArgumentParser() parser.add_argument( "--world_size", type=int, default=1, help="world size, only support tensor parallelism now", ) parser.add_argument("--tp_size", type=int, default=1) parser.add_argument("--pp_size", type=int, default=1) parser.add_argument("--hf_model_dir", type=str, default=None) parser.add_argument("--version", "-v", type=str, default="1", help="qwen version, support 1, 1.5") parser.add_argument("--ft_dir_path", type=str, default=None) parser.add_argument( "--dtype", type=str, default="float16", choices=["float32", "bfloat16", "float16"], ) parser.add_argument( '--timing_cache', type=str, default='model.cache', help= 'The path of to read timing cache from, will be ignored if the file does not exist' ) parser.add_argument('--log_level', type=str, default='info', choices=[ 'internal_error', 'error', 'warning', 'info', 'verbose', ]) parser.add_argument('--vocab_size', type=int, default=32000) parser.add_argument('--n_layer', type=int, default=32) parser.add_argument('--n_positions', type=int, default=2048) parser.add_argument('--n_embd', type=int, default=4096) parser.add_argument('--n_head', type=int, default=32) parser.add_argument('--n_kv_head', type=int, default=None) parser.add_argument('--inter_size', type=int, default=11008) parser.add_argument('--hidden_act', type=str, default='silu') parser.add_argument('--max_batch_size', type=int, default=2) parser.add_argument('--max_input_len', type=int, default=2048) parser.add_argument('--max_output_len', type=int, default=2048) parser.add_argument('--max_beam_width', type=int, default=1) parser.add_argument('--rotary_base', type=float, default=10000.0) parser.add_argument('--rotary_scaling', nargs=2, type=str, default=None) parser.add_argument('--use_gpt_attention_plugin', nargs='?', type=str, default="float16", choices=['float16', 'bfloat16', 'float32', None]) parser.add_argument('--use_gemm_plugin', nargs='?', type=str, default="float16", choices=['float16', 'bfloat16', 'float32', None]) parser.add_argument('--parallel_build', default=False, action='store_true') parser.add_argument('--enable_context_fmha', default=False, action='store_true') parser.add_argument('--enable_context_fmha_fp32_acc', default=False, action='store_true') parser.add_argument('--visualize', default=False, action='store_true') parser.add_argument('--enable_debug_output', default=False, action='store_true') parser.add_argument('--gpus_per_node', type=int, default=8) parser.add_argument('--builder_opt', type=int, default=None) parser.add_argument( '--output_dir', type=str, default='engine_outputs', help= 'The path to save the serialized engine files, timing cache file and model configs' ) parser.add_argument('--remove_input_padding', default=False, action='store_true') # Arguments related to the quantization of the model. parser.add_argument( '--use_smooth_quant', default=False, action="store_true", help= 'Use the SmoothQuant method to quantize activations and weights for the various GEMMs.' 'See --per_channel and --per_token for finer-grained quantization options.' ) 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( '--per_group', default=False, action="store_true", help= 'By default, we use a single static scaling factor to scale weights in the int4 range. ' 'per_group chooses at run time, and for each group, a custom scaling factor. ' 'The flag is built for GPTQ/AWQ quantization.') 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'], 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( '--use_inflight_batching', action="store_true", default=False, help="Activates inflight batching mode of gptAttentionPlugin.") parser.add_argument( '--paged_kv_cache', action="store_true", default=False, help= 'By default we use contiguous KV cache. By setting this flag you enable paged KV cache' ) parser.add_argument('--tokens_per_block', type=int, default=128, help='Number of tokens per block in paged KV cache') parser.add_argument( '--max_num_tokens', type=int, default=None, help='Define the max number of tokens supported by the engine') 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' ) parser.add_argument( '--use_parallel_embedding', action="store_true", default=False, help= 'By default embedding parallelism is disabled. By setting this flag, embedding parallelism is enabled' ) parser.add_argument( '--embedding_sharding_dim', type=int, default=1, # Meta does TP on hidden dim choices=[0, 1], help= 'By default the embedding lookup table is sharded along vocab dimension (embedding_sharding_dim=0). ' 'To shard it along hidden dimension, set embedding_sharding_dim=1' 'Note: embedding sharing is only enabled when embedding_sharding_dim = 0' ) parser.add_argument( '--strongly_typed', default=False, action="store_true", help= 'This option is introduced with trt 9.1.0.1+ and will reduce the building time significantly for fp8.' ) parser.add_argument( '--opt_memory_use', default=False, action="store_true", help='Whether to use Host memory optimization for building engine') parser.add_argument( '--use_custom_all_reduce', action='store_true', help= 'Activates latency-optimized algorithm for all-reduce instead of NCCL.') parser.add_argument('--gather_all_token_logits', action='store_true', default=False) args = parser.parse_args() assert not ( args.use_smooth_quant and args.use_weight_only ), "You cannot enable both SmoothQuant and INT8 weight-only together." if not args.remove_input_padding: if args.use_gpt_attention_plugin: logger.warning( f"It is recommended to specify --remove_input_padding when using GPT attention plugin" ) if args.use_inflight_batching: if not args.use_gpt_attention_plugin: args.use_gpt_attention_plugin = 'float16' logger.info( f"Using GPT attention plugin for inflight batching mode. Setting to default '{args.use_gpt_attention_plugin}'" ) if not args.remove_input_padding: args.remove_input_padding = True logger.info( "Using remove input padding for inflight batching mode.") if not args.paged_kv_cache: args.paged_kv_cache = True logger.info("Using paged KV cache for inflight batching mode.") if args.use_smooth_quant: args.quant_mode = QuantMode.use_smooth_quant(args.per_token, args.per_channel) elif args.use_weight_only: if args.per_group: args.quant_mode = QuantMode.from_description( quantize_weights=True, quantize_activations=False, per_token=False, per_channel=False, per_group=True, use_int4_weights=True) else: args.quant_mode = QuantMode.use_weight_only( args.weight_only_precision == 'int4') else: args.quant_mode = QuantMode(0) if args.int8_kv_cache: args.quant_mode = args.quant_mode.set_int8_kv_cache() if args.hf_model_dir is not None: hf_config = AutoConfig.from_pretrained( args.hf_model_dir, trust_remote_code=True, ) args.inter_size = hf_config.intermediate_size # override the inter_size for QWen args.n_embd = hf_config.hidden_size args.n_head = hf_config.num_attention_heads if hasattr(hf_config, "num_key_value_heads"): args.n_kv_head = hf_config.num_key_value_heads args.n_layer = hf_config.num_hidden_layers args.n_positions = hf_config.max_position_embeddings args.vocab_size = hf_config.vocab_size args.hidden_act = "silu" if hasattr(hf_config, "kv_channels"): args.kv_channels = hf_config.kv_channels elif hasattr(hf_config, "num_key_value_heads"): args.kv_channels = hf_config.num_key_value_heads else: raise if hasattr(hf_config, "rotary_emb_base"): args.rotary_emb_base = hf_config.rotary_emb_base else: args.rotary_emb_base = 10000.0 assert args.use_gpt_attention_plugin is not None, "QWen must use gpt attention plugin" # if args.n_kv_head is not None and args.n_kv_head != args.n_head: # assert (args.n_head % args.n_kv_head) == 0, \ # "MQA/GQA requires the number of heads to be divisible by the number of K/V heads." # assert args.n_kv_head == args.tp_size, \ # "The current implementation of GQA requires the number of K/V heads to match the number of GPUs." \ # "This limitation will be removed in a future version." assert args.pp_size * args.tp_size == args.world_size if args.max_num_tokens is not None: assert args.enable_context_fmha assert (math.log2(args.tokens_per_block).is_integer() ), "tokens_per_block must be power of 2" if args.enable_context_fmha or args.enable_context_fmha_fp32_acc: assert (args.tokens_per_block >= 128), "Context fMHA requires >= 128 tokens per block" return args def build_rank_engine(builder: Builder, builder_config: xtrt_llm.builder.BuilderConfig, engine_name, rank, multi_query_mode, args): ''' @brief: Build the engine on the given rank. @param rank: The rank to build the engine. @param args: The cmd line arguments. @return: The built engine. ''' kv_dtype = str_dtype_to_xtrt(args.dtype) mapping = Mapping(world_size=args.world_size, rank=rank, tp_size=args.tp_size, pp_size=args.pp_size) # Initialize Module assert args.version in ["1", "1.5"], "Only support version 1 and 1.5" if args.version == "1.5": from qwen2_weight import load_from_ft, load_from_hf_qwen xtrt_llm_qwen = xtrt_llm.models.Qwen2ForCausalLM( num_layers=args.n_layer, num_heads=args.n_head, num_kv_heads=args.n_kv_head, hidden_size=args.n_embd, seq_length=args.max_input_len, vocab_size=args.vocab_size, hidden_act=args.hidden_act, max_position_embeddings=args.n_positions, dtype=kv_dtype, mlp_hidden_size=args.inter_size, mapping=mapping, rotary_base=args.rotary_base, rotary_scaling=args.rotary_scaling, use_parallel_embedding=args.use_parallel_embedding, embedding_sharding_dim=args.embedding_sharding_dim, quant_mode=args.quant_mode, gather_all_token_logits=args.gather_all_token_logits, ) else: from qwen_weight import load_from_ft, load_from_hf_qwen xtrt_llm_qwen = xtrt_llm.models.QWenForCausalLM( num_layers=args.n_layer, num_heads=args.n_head, num_kv_heads=args.n_kv_head, hidden_size=args.n_embd, seq_length=args.max_input_len, vocab_size=args.vocab_size, hidden_act=args.hidden_act, max_position_embeddings=args.n_positions, dtype=kv_dtype, mlp_hidden_size=args.inter_size, neox_rotary_style=True, mapping=mapping, rotary_base=args.rotary_base, rotary_scaling=args.rotary_scaling, use_parallel_embedding=args.use_parallel_embedding, embedding_sharding_dim=args.embedding_sharding_dim, quant_mode=args.quant_mode, gather_all_token_logits=args.gather_all_token_logits, ) quantize_kwargs = {} if args.use_smooth_quant or args.use_weight_only: if args.weight_only_precision == 'int4_awq': quantize_kwargs = { "group_size": args.group_size, "zero": False, "pre_quant_scale": True, "exclude_modules": [], } elif args.weight_only_precision == 'int4_gptq': quantize_kwargs = { "group_size": args.group_size, "zero": True, "pre_quant_scale": False, } xtrt_llm_qwen = quantize_model(xtrt_llm_qwen, args.quant_mode, **quantize_kwargs) ft_dir_path = args.ft_dir_path if args.hf_model_dir is not None and \ (ft_dir_path is None or not os.path.exists(ft_dir_path)): logger.info(f'Loading HF QWen ... from {args.hf_model_dir}') tik = time.time() if args.version == "1": hf_qwen = AutoModelForCausalLM.from_pretrained( args.hf_model_dir, device_map={ "transformer": "cpu", "lm_head": "cpu", }, # Load to CPU memory torch_dtype="auto", trust_remote_code=True, ) else: hf_qwen = Qwen2ForCausalLM.from_pretrained( args.hf_model_dir, # device_map="cpu", device_map={ "model": "cpu", "lm_head": "cpu" }, # Load to CPU memory torch_dtype="auto", trust_remote_code=True, ) tok = time.time() t = time.strftime('%H:%M:%S', time.gmtime(tok - tik)) logger.info(f'HF QWen loaded. Total time: {t}') load_from_hf_qwen(xtrt_llm_qwen, hf_qwen, mapping, max_position_embeddings=args.n_positions, kv_channels=args.kv_channels, rotary_emb_base=args.rotary_emb_base, dtype=args.dtype, multi_query_mode=multi_query_mode) del hf_qwen elif ft_dir_path is not None: dir_path = ft_dir_path logger.info(f'Loading FT QWen ... from {ft_dir_path}') load_from_ft(xtrt_llm_qwen, dir_path, mapping, dtype=args.dtype, multi_query_mode=multi_query_mode) else: raise ValueError( "You must specify either --hf_model_dir or --ft_dir_path") # Module -> Network network = builder.create_network() network.trt_network.name = engine_name if args.use_gpt_attention_plugin: network.plugin_config.set_gpt_attention_plugin( dtype=args.use_gpt_attention_plugin) if args.use_gemm_plugin: network.plugin_config.set_gemm_plugin(dtype=args.use_gemm_plugin) # Quantization plugins. if args.use_smooth_quant: network.plugin_config.set_smooth_quant_gemm_plugin(dtype=args.dtype) network.plugin_config.set_rmsnorm_quantization_plugin(dtype=args.dtype) network.plugin_config.set_quantize_tensor_plugin() network.plugin_config.set_quantize_per_token_plugin() assert not (args.enable_context_fmha and args.enable_context_fmha_fp32_acc) if args.enable_context_fmha: network.plugin_config.set_context_fmha(ContextFMHAType.enabled) if args.enable_context_fmha_fp32_acc: network.plugin_config.set_context_fmha( ContextFMHAType.enabled_with_fp32_acc) if args.use_weight_only: if args.per_group: network.plugin_config.set_weight_only_groupwise_quant_matmul_plugin( dtype='float16') else: network.plugin_config.set_weight_only_quant_matmul_plugin( dtype='float16') if args.quant_mode.is_weight_only(): builder_config.trt_builder_config.use_weight_only = args.weight_only_precision if args.world_size > 1: network.plugin_config.set_nccl_plugin(args.dtype, args.use_custom_all_reduce) if args.remove_input_padding: network.plugin_config.enable_remove_input_padding() if args.paged_kv_cache: network.plugin_config.enable_paged_kv_cache(args.tokens_per_block) with net_guard(network): # Prepare network.set_named_parameters(xtrt_llm_qwen.named_parameters()) # Forward inputs = xtrt_llm_qwen.prepare_inputs( max_batch_size=args.max_batch_size, max_input_len=args.max_input_len, max_new_tokens=args.max_output_len, use_cache=True, max_beam_width=args.max_beam_width, max_num_tokens=args.max_num_tokens, ) xtrt_llm_qwen(*inputs) if args.enable_debug_output: # mark intermediate nodes' outputs for k, v in xtrt_llm_qwen.named_network_outputs(): v = v.trt_tensor v.name = k network.trt_network.mark_output(v) v.dtype = kv_dtype if args.visualize: model_path = os.path.join(args.output_dir, 'test.onnx') to_onnx(network.trt_network, model_path) engine = None # Network -> Engine engine = builder.build_engine(network, builder_config) if rank == 0: config_path = os.path.join(args.output_dir, 'config.json') builder.save_config(builder_config, config_path) if args.opt_memory_use: return engine, network return engine def build(rank, args): torch.cuda.set_device(rank % args.gpus_per_node) xtrt_llm.logger.set_level(args.log_level) if not os.path.exists(args.output_dir): os.makedirs(args.output_dir) multi_query_mode = (args.n_kv_head is not None) and (args.n_kv_head != args.n_head) # when doing serializing build, all ranks share one engine builder = Builder() cache = None for cur_rank in range(args.world_size): # skip other ranks if parallel_build is enabled if args.parallel_build and cur_rank != rank: continue int8_trt_flag = args.quant_mode.has_act_and_weight_quant() or ( not args.paged_kv_cache and args.quant_mode.has_int8_kv_cache()) builder_config = builder.create_builder_config( name=MODEL_NAME, precision=args.dtype, timing_cache=args.timing_cache if cache is None else cache, tensor_parallel=args.tp_size, pipeline_parallel=args.pp_size, parallel_build=args.parallel_build, num_layers=args.n_layer, num_heads=args.n_head, hidden_size=args.n_embd, inter_size=args.inter_size, vocab_size=args.vocab_size, hidden_act=args.hidden_act, max_position_embeddings=args.n_positions, max_batch_size=args.max_batch_size, max_beam_width=args.max_beam_width, max_input_len=args.max_input_len, max_output_len=args.max_output_len, max_num_tokens=args.max_num_tokens, fusion_pattern_list=["remove_dup_mask"], int8=int8_trt_flag, fp8=args.quant_mode.has_fp8_qdq(), quant_mode=args.quant_mode, strongly_typed=args.strongly_typed, opt_level=args.builder_opt, max_prompt_embedding_table_size=0, # max_prompt_embedding_table_size=args.max_prompt_embedding_table_size, gather_all_token_logits=args.gather_all_token_logits) guard = xtrt_llm.fusion_patterns.FuseonPatternGuard() print(guard) engine_name = get_engine_name(MODEL_NAME, args.dtype, args.tp_size, args.pp_size, cur_rank) if args.opt_memory_use: engine, network = build_rank_engine(builder, builder_config, engine_name, cur_rank, multi_query_mode, args) else: engine = build_rank_engine(builder, builder_config, engine_name, cur_rank, multi_query_mode, args) assert engine is not None, f'Failed to build engine for rank {cur_rank}' if cur_rank == 0: # Use in-memory timing cache for multiple builder passes. if not args.parallel_build: cache = builder_config.trt_builder_config.get_timing_cache() serialize_engine(engine, os.path.join(args.output_dir, engine_name)) del engine if args.opt_memory_use: network.__del__() # if rank == 0: # ok = builder.save_timing_cache( # builder_config, os.path.join(args.output_dir, "model.cache")) # assert ok, "Failed to save timing cache." if __name__ == '__main__': args = parse_arguments() logger.set_level(args.log_level) tik = time.time() if args.version == "1.5": MODEL_NAME = 'qwen2' if args.parallel_build and args.world_size > 1 and \ torch.cuda.device_count() >= args.world_size: logger.warning( f'Parallelly build TensorRT engines. Please make sure that all of the {args.world_size} GPUs are totally free.' ) mp.spawn(build, nprocs=args.world_size, args=(args, )) else: args.parallel_build = False logger.info('Serially build TensorRT engines.') build(0, args) tok = time.time() t = time.strftime('%H:%M:%S', time.gmtime(tok - tik)) logger.info(f'Total time of building all {args.world_size} engines: {t}')