[Bugfix] Fix top logprobs token placeholders in /inference/v1/generate (#42887)

Signed-off-by: Sage Ahrac <sagiahrak@gmail.com>
This commit is contained in:
Sage
2026-05-19 10:21:49 +00:00
committed by GitHub
parent 129019f334
commit a78b842d0e
2 changed files with 33 additions and 3 deletions
@@ -0,0 +1,28 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from vllm.entrypoints.serve.disagg.serving import ServingTokens
from vllm.logprobs import Logprob
def test_top_logprobs_alternatives_have_own_token_ids():
"""Each top_logprobs alternative must carry its own token_id placeholder."""
result = ServingTokens._create_tokens_logprobs(
None,
token_ids=[262],
top_logprobs=[{262: Logprob(-0.1), 257: Logprob(-1.2), 428: Logprob(-2.3)}],
num_output_top_logprobs=2,
)
tokens = {e.token for e in result.content[0].top_logprobs}
assert tokens == {"token_id:262", "token_id:257"}, f"got {tokens}"
def test_logprobs_zero_emits_sampled_token():
"""logprobs=0 must still emit 1 entry (the sampled token)."""
result = ServingTokens._create_tokens_logprobs(
None,
token_ids=[7],
top_logprobs=[{7: Logprob(-0.9), 8: Logprob(-1.1)}],
num_output_top_logprobs=0,
)
assert len(result.content[0].top_logprobs) == 1
+5 -3
View File
@@ -460,10 +460,12 @@ class ServingTokens(OpenAIServing):
logprob=max(step_token.logprob, -9999.0),
top_logprobs=[
ChatCompletionLogProb(
token=token,
logprob=max(p[1].logprob, -9999.0),
token=f"token_id:{token_id}",
logprob=max(logprob.logprob, -9999.0),
)
for i, (token_id, logprob) in enumerate(
step_top_logprobs.items()
)
for i, p in enumerate(step_top_logprobs.items())
if num_output_top_logprobs is not None
and i < max(num_output_top_logprobs, 1)
],