biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[CI/Test] improve robustness of test (vllm_runner) (#5357)

Browse Source 
[CI/Test] improve robustness of test by replacing del with context manager (vllm_runner) (#5357)
This commit is contained in:
youkaichao
2024-06-08 01:59:20 -07:00
committed by GitHub
parent 9fb900f90c
commit 8ea5e44a43
28 changed files with 431 additions and 470 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
tests/basic_correctness/test_chunked_prefill.py
View File

@@ -43,17 +43,16 @@ def test_models(
with hf_runner(model, dtype=dtype) as hf_model:
hf_outputs = hf_model.generate_greedy(example_prompts, max_tokens)
vllm_model = vllm_runner(
model,
dtype=dtype,
max_num_batched_tokens=max_num_batched_tokens,
enable_chunked_prefill=enable_chunked_prefill,
tensor_parallel_size=tensor_parallel_size,
enforce_eager=enforce_eager,
max_num_seqs=max_num_seqs,
)
vllm_outputs = vllm_model.generate_greedy(example_prompts, max_tokens)
del vllm_model
with vllm_runner(
model,
dtype=dtype,
max_num_batched_tokens=max_num_batched_tokens,
enable_chunked_prefill=enable_chunked_prefill,
tensor_parallel_size=tensor_parallel_size,
enforce_eager=enforce_eager,
max_num_seqs=max_num_seqs,
) as vllm_model:
vllm_outputs = vllm_model.generate_greedy(example_prompts, max_tokens)
for i in range(len(example_prompts)):
hf_output_ids, hf_output_str = hf_outputs[i]