"""Benchmark offline inference throughput.""" import argparse import json import random import time from typing import List, Optional, Tuple import torch from transformers import AutoModelForCausalLM, PreTrainedTokenizerBase from tqdm import tqdm from aphrodite import LLM, SamplingParams from aphrodite.transformers_utils.tokenizer import get_tokenizer def sample_requests( dataset_path: str, num_requests: int, tokenizer: PreTrainedTokenizerBase, ) -> List[Tuple[str, int, int]]: # Load the dataset. with open(dataset_path) as f: dataset = json.load(f) # Filter out the conversations with less than 2 turns. dataset = [data for data in dataset if len(data["conversations"]) >= 2] # Only keep the first two turns of each conversation. dataset = [(data["conversations"][0]["value"], data["conversations"][1]["value"]) for data in dataset] # Tokenize the prompts and completions. prompts = [prompt for prompt, _ in dataset] prompt_token_ids = tokenizer(prompts).input_ids completions = [completion for _, completion in dataset] completion_token_ids = tokenizer(completions).input_ids tokenized_dataset = [] for i in range(len(dataset)): output_len = len(completion_token_ids[i]) tokenized_dataset.append((prompts[i], prompt_token_ids[i], output_len)) # Filter out too long sequences. filtered_dataset: List[Tuple[str, int, int]] = [] for prompt, prompt_token_ids, output_len in tokenized_dataset: prompt_len = len(prompt_token_ids) if prompt_len < 4 or output_len < 4: # Prune too short sequences. continue if prompt_len > 1024 or prompt_len + output_len > 2048: # Prune too long sequences. continue filtered_dataset.append((prompt, prompt_len, output_len)) # Sample the requests. sampled_requests = random.sample(filtered_dataset, num_requests) return sampled_requests def run_aphrodite( requests: List[Tuple[str, int, int]], model: str, tokenizer: str, quantization: Optional[str], tensor_parallel_size: int, seed: int, n: int, use_beam_search: bool, trust_remote_code: bool, dtype: str, kv_cache_dtype: str, disable_custom_all_reduce: bool, context_shift: bool, enforce_eager: bool, enable_chunked_prefill: bool, max_num_batched_tokens: int, speculative_model: Optional[str] = None, num_speculative_tokens: Optional[int] = None, use_v2_block_manager: bool = False, ) -> float: llm = LLM( model=model, tokenizer=tokenizer, quantization=quantization, tensor_parallel_size=tensor_parallel_size, seed=seed, trust_remote_code=trust_remote_code, dtype=dtype, kv_cache_dtype=kv_cache_dtype, disable_custom_all_reduce=disable_custom_all_reduce, context_shift=context_shift, enforce_eager=enforce_eager, enable_chunked_prefill=enable_chunked_prefill, max_num_batched_tokens=max_num_batched_tokens, speculative_model=speculative_model, num_speculative_tokens=num_speculative_tokens, use_v2_block_manager=use_v2_block_manager, ) # Add the requests to the engine. for prompt, _, output_len in requests: sampling_params = SamplingParams( n=n, temperature=0.0 if use_beam_search else 1.0, top_p=1.0, use_beam_search=use_beam_search, ignore_eos=True, max_tokens=output_len, ) # FIXME: Do not use internal method. llm._add_request( # pylint: disable=protected-access prompt=prompt, prompt_token_ids=None, sampling_params=sampling_params, ) start = time.perf_counter() # FIXME Do use internal method. llm._run_engine(use_tqdm=True) # pylint: disable=protected-access end = time.perf_counter() return end - start def run_hf( requests: List[Tuple[str, int, int]], model: str, tokenizer: PreTrainedTokenizerBase, n: int, use_beam_search: bool, max_batch_size: int, trust_remote_code: bool, ) -> float: assert not use_beam_search llm = AutoModelForCausalLM.from_pretrained( model, torch_dtype=torch.float16, trust_remote_code=trust_remote_code) if llm.config.model_type == "llama": # To enable padding in the HF backend. tokenizer.pad_token = tokenizer.eos_token llm = llm.cuda() pbar = tqdm(total=len(requests)) start = time.perf_counter() batch: List[str] = [] max_prompt_len = 0 max_output_len = 0 for i in range(len(requests)): prompt, prompt_len, output_len = requests[i] # Add the prompt to the batch. batch.append(prompt) max_prompt_len = max(max_prompt_len, prompt_len) max_output_len = max(max_output_len, output_len) if len(batch) < max_batch_size and i != len(requests) - 1: # Check if we can add more requests to the batch. _, next_prompt_len, next_output_len = requests[i + 1] if (max(max_prompt_len, next_prompt_len) + max(max_output_len, next_output_len)) <= 2048: # We can add more requests to the batch. continue # Generate the sequences. input_ids = tokenizer(batch, return_tensors="pt", padding=True).input_ids llm_outputs = llm.generate( input_ids=input_ids.cuda(), do_sample=not use_beam_search, num_return_sequences=n, temperature=1.0, top_p=1.0, use_cache=True, max_new_tokens=max_output_len, ) # Include the decoding time. tokenizer.batch_decode(llm_outputs, skip_special_tokens=True) pbar.update(len(batch)) # Clear the batch. batch = [] max_prompt_len = 0 max_output_len = 0 end = time.perf_counter() return end - start def main(args: argparse.Namespace): # pylint: disable=redefined-outer-name print(args) random.seed(args.seed) # Sample the requests. tokenizer = get_tokenizer(args.tokenizer, trust_remote_code=args.trust_remote_code) requests = sample_requests(args.dataset, args.num_prompts, tokenizer) if args.backend == "aphrodite": elapsed_time = run_aphrodite( requests, args.model, args.tokenizer, args.quantization, args.tensor_parallel_size, args.seed, args.n, args.use_beam_search, args.trust_remote_code, args.dtype, args.kv_cache_dtype, args.disable_custom_all_reduce, args.context_shift, args.enforce_eager, args.enable_chunked_prefill, args.max_num_batched_tokens) elif args.backend == "hf": assert args.tensor_parallel_size == 1 elapsed_time = run_hf(requests, args.model, tokenizer, args.n, args.use_beam_search, args.hf_max_batch_size, args.trust_remote_code) else: raise ValueError(f"Unknown backend: {args.backend}") total_input_tokens = sum(prompt_len for _, prompt_len, _ in requests) total_output_tokens = sum(output_len for _, _, output_len in requests) print(f"Throughput: {len(requests) / elapsed_time:.2f} requests/s, " f"Input tokens/s: {total_input_tokens / elapsed_time:.2f}, " f"Output tokens/s: {total_output_tokens / elapsed_time:.2f}") if __name__ == "__main__": parser = argparse.ArgumentParser(description="Benchmark the throughput.") parser.add_argument("--backend", type=str, choices=["aphrodite", "hf"], default="aphrodite") parser.add_argument("--dataset", type=str, required=True, help="Path to the dataset.") parser.add_argument("--model", type=str, default="EleutherAI/pythia-70m-deduped") parser.add_argument("--tokenizer", type=str, default=None) parser.add_argument( "--quantization", "-q", choices=["awq", "gguf", "bnb", "gptq", "squeezellm", "marlin", None], default=None) parser.add_argument("--gpu-memory-utilization", type=float, default=0.88) parser.add_argument("--tensor-parallel-size", "-tp", type=int, default=1) parser.add_argument("--n", type=int, default=1, help="Number of generated sequences per prompt.") parser.add_argument("--use-beam-search", action="store_true") parser.add_argument("--num-prompts", type=int, default=1000, help="Number of prompts to process.") parser.add_argument("--seed", type=int, default=0) parser.add_argument("--hf-max-batch-size", type=int, default=None, help="Maximum batch size for HF backend.") parser.add_argument("--trust-remote-code", action="store_true", help="trust remote code from huggingface") parser.add_argument( "--dtype", type=str, default="auto", choices=["auto", "half", "float16", "bfloat16", "float", "float32"], help="data type for model weights and activations. " "The 'auto' option will use FP16 precision " "for FP32 and FP16 models, and BF16 precision " "for BF16 models.") parser.add_argument("--kv-cache-dtype", type=str, default="auto", choices=["auto", "fp8_e5m2"], help="The Data Type for the KV cache.") parser.add_argument( "--disable-custom-all-reduce", action="store_true", help="disable custom all reduce for the Aphrodite backend") parser.add_argument( "--context-shift", action="store_true", help="enable context shifting for the Aphrodite backend") parser.add_argument("--enforce-eager", type=lambda x: (str(x).lower() == 'true'), default=True, help="enforce eager mode for the Aphrodite backend") parser.add_argument( "--enable-chunked-prefill", action="store_true", help="enable chunked prefill for the Aphrodite backend") parser.add_argument("--max-num-batched-tokens", type=int, help="maximum number of batched tokens for the " "Aphrodite backend") parser.add_argument("--speculative-model", type=str, help="speculative model for the Aphrodite backend") parser.add_argument("--num-speculative-tokens", type=int, help="number of speculative tokens for the " "Aphrodite backend") parser.add_argument("--use-v2-block-manager", action="store_true", help="use v2 block manager for the Aphrodite backend") args = parser.parse_args() if args.backend == "aphrodite": if args.hf_max_batch_size is not None: raise ValueError("HF max batch size is only for HF backend.") elif args.backend == "hf": if args.hf_max_batch_size is None: raise ValueError("HF max batch size is required for HF backend.") if args.quantization is not None: raise ValueError("Quantization is only for aphrodite backend.") if args.tokenizer is None: args.tokenizer = args.model main(args)