[Core] [Bugfix] [Multimodal] Fix multimodal profiling and generation for SFT/PTQed models (#20058)

Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
2025-06-30 13:26:49 -04:00
parent 551ef1631a
commit d8cf819a9a
41 changed files with 207 additions and 38 deletions
--- a/vllm/model_executor/models/granite_speech.py
+++ b/vllm/model_executor/models/granite_speech.py
@@ -141,6 +141,7 @@ class GraniteSpeechMultiModalProcessor(
        prompt: str,
        mm_data: Mapping[str, object],
        mm_kwargs: Mapping[str, object],
+        tok_kwargs: Mapping[str, object],
    ) -> BatchFeature:
        mm_data = dict(mm_data)
        audios = mm_data.pop("audios", [])
@@ -153,6 +154,7 @@ class GraniteSpeechMultiModalProcessor(
            prompt=prompt,
            mm_data=mm_data,
            mm_kwargs=mm_kwargs,
+            tok_kwargs=tok_kwargs,
        )

        if "audio" in mm_data: