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

[Core] Switch Flat logprob control from environment variable to SamplingParams (#28914)

Browse Source 
Signed-off-by: Jialin Ouyang <Jialin.Ouyang@gmail.com>
Co-authored-by: 22quinn <33176974+22quinn@users.noreply.github.com>
This commit is contained in:
Jialin Ouyang
2025-11-18 18:10:02 -08:00
committed by GitHub
parent da94c7c0eb
commit 40b6b38f2c
6 changed files with 33 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
tests/samplers/test_logprobs.py
View File

@@ -24,9 +24,7 @@ def test_ranks(
greedy,
flat_logprobs,
example_prompts,
monkeypatch: pytest.MonkeyPatch,
):
monkeypatch.setenv("VLLM_FLAT_LOGPROBS", "1" if flat_logprobs else "0")
with vllm_runner(model, dtype=dtype, max_logprobs=MAX_LOGPROBS) as vllm_model:
tokenizer = vllm_model.llm.get_tokenizer()
example_prompt_tokens = [tokenizer.encode(prompt) for prompt in example_prompts]
@@ -36,6 +34,7 @@ def test_ranks(
max_tokens=MAX_TOKENS,
logprobs=NUM_TOP_LOGPROBS,
prompt_logprobs=NUM_PROMPT_LOGPROBS,
flat_logprobs=flat_logprobs,
)
results = vllm_model.generate_w_logprobs(example_prompts, sampling_params)