diff --git a/tests/entrypoints/serve/disagg/test_tokens_logprobs.py b/tests/entrypoints/serve/disagg/test_tokens_logprobs.py new file mode 100644 index 0000000000..844dd24d54 --- /dev/null +++ b/tests/entrypoints/serve/disagg/test_tokens_logprobs.py @@ -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 diff --git a/vllm/entrypoints/serve/disagg/serving.py b/vllm/entrypoints/serve/disagg/serving.py index 0edfa61487..30126df4b7 100644 --- a/vllm/entrypoints/serve/disagg/serving.py +++ b/vllm/entrypoints/serve/disagg/serving.py @@ -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) ],