[Bugfix][V1/V2] Fix prompt_logprobs to respect logprobs_mode (#47680)

Signed-off-by: Wojciech Wais <wojciech.wais@gmail.com>
Signed-off-by: Federico Kamelhar <209537060+fede-kamel@users.noreply.github.com>
Signed-off-by: Allen Shen <aoshen@inferact.ai>
Co-authored-by: Wojciech Wais <wojciech.wais@gmail.com>
Co-authored-by: Federico Kamelhar <209537060+fede-kamel@users.noreply.github.com>
This commit is contained in:
aoshen02
2026-07-17 21:58:59 +01:00
committed by GitHub
co-authored by Wojciech Wais Federico Kamelhar
parent 088c0be268
commit 41ea2dd44a
11 changed files with 106 additions and 40 deletions
+2
View File
@@ -233,6 +233,8 @@ class ModelConfig:
Raw means the values before applying any logit processors, like bad words.
Processed means the values after applying all processors, including
temperature and top_k/top_p.
Note: for prompt_logprobs, processed_* and raw_* yield identical results
because prompt tokens do not go through sampling processors.
"""
use_fp64_gumbel: bool = False
"""Whether to use FP64 (instead of FP32) random noise for Gumbel-max and
-7
View File
@@ -2202,13 +2202,6 @@ class VllmConfig:
if model_config is not None and model_config.enable_prompt_embeds:
unsupported.append("prompt embeds")
if (
model_config is not None
and model_config.runner_type == "generate"
and model_config.logprobs_mode in ("raw_logits", "processed_logits")
):
unsupported.append(f"logprobs mode '{model_config.logprobs_mode}'")
if self.cache_config.kv_sharing_fast_prefill:
# Will be added by https://github.com/vllm-project/vllm/pull/35045
unsupported.append("KV sharing fast prefill")