Benchmarking script for openai chat completion api are not supported

[wangchen615]

When running vllm with openai chat apis, the benchmarking script will fail as it asserts the backend API of assert api_url.endswith("v1/completions").

python benchmark_serving.py --backend openai --model mistralai/Mistral-7B-v0.1 --dataset ShareGPT_V3_unfiltered_cleaned_split.json --save-result

The logs are as follows:

Namespace(backend='openai', version='N/A', base_url=None, host='localhost', port=8000, endpoint='/generate', dataset='ShareGPT_V3_unfiltered_cleaned_split.json', model='mistralai/Mistral-7B-v0.1', tokenizer=None, best_of=1, use_beam_search=False, num_prompts=1000, request_rate=inf, seed=0, trust_remote_code=False, disable_tqdm=False, save_result=True)
  0%|                                                      | 0/1000 [00:00<?, ?it/s]Traffic request rate: inf
Traceback (most recent call last):
  File "/home/chenw/vllm/benchmarks/benchmark_serving.py", line 387, in <module>
    main(args)
  File "/home/chenw/vllm/benchmarks/benchmark_serving.py", line 259, in main
    benchmark_result = asyncio.run(
  File "/home/chenw/miniconda3/envs/myenv/lib/python3.9/asyncio/runners.py", line 44, in run
    return loop.run_until_complete(main)
  File "/home/chenw/miniconda3/envs/myenv/lib/python3.9/asyncio/base_events.py", line 647, in run_until_complete
    return future.result()
  File "/home/chenw/vllm/benchmarks/benchmark_serving.py", line 195, in benchmark
    outputs = await asyncio.gather(*tasks)
  File "/home/chenw/vllm/benchmarks/backend_request_func.py", line 223, in async_request_openai_completions
    assert api_url.endswith("v1/completions")
AssertionError
  0%|                                                      | 0/1000 [00:00<?, ?it/s]

The backend_request_func.py should not only allow chat apis like: assert api_url.endswith("v1/chat/completions").