[CI/Test] improve robustness of test (hf_runner) (#5347)

[CI/Test] improve robustness of test by replacing del with context manager (hf_runner) (#5347)
This commit is contained in:
youkaichao
2024-06-07 22:31:32 -07:00
committed by GitHub
parent c96fc06747
commit 9fb900f90c
14 changed files with 48 additions and 61 deletions
+3 -4
View File
@@ -30,10 +30,9 @@ def test_beam_search_single_input(
beam_width: int,
) -> None:
example_prompts = example_prompts[:1]
hf_model = hf_runner(model, dtype=dtype)
hf_outputs = hf_model.generate_beam_search(example_prompts, beam_width,
max_tokens)
del hf_model
with hf_runner(model, dtype=dtype) as hf_model:
hf_outputs = hf_model.generate_beam_search(example_prompts, beam_width,
max_tokens)
vllm_model = vllm_runner(model, dtype=dtype)
vllm_outputs = vllm_model.generate_beam_search(example_prompts, beam_width,
+5 -6
View File
@@ -32,12 +32,11 @@ def test_get_prompt_logprobs(
max_num_batched_tokens = chunked_prefill_token_size
max_tokens = 5
hf_model = hf_runner(model, dtype=dtype)
hf_logprobs = hf_model.generate_greedy_logprobs(
example_prompts,
max_tokens=max_tokens,
)
del hf_model
with hf_runner(model, dtype=dtype) as hf_model:
hf_logprobs = hf_model.generate_greedy_logprobs(
example_prompts,
max_tokens=max_tokens,
)
vllm_model = vllm_runner(
model,