Allow untokenized output from apply_transcription_request (#49944)

* Allow untokenized output from `apply_transcription_request`

Add a `tokenize` argument to the chat-template-based ASR processors so
that inference engines, which run the processor themselves, can get the
formatted prompts. Also make VibeVoice-ASR's `extract_speaker_dict` keep
the complete segments of truncated output instead of raising.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

* doc nit

---------

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: eustlb <94853470+eustlb@users.noreply.github.com>

Author: 3 people

src/transformers/models/audioflamingo3/processing_audioflamingo3.py

@@ -183,8 +183,9 @@ def apply_transcription_request(
         self,
         audio: str | list[str] | AudioInput,
         prompt: str | list[str] | None = None,
+        tokenize: bool = True,
         **kwargs: Unpack[AudioFlamingo3ProcessorKwargs],
-    ) -> BatchFeature:
+    ) -> BatchFeature | list[str]:
         """
         Prepare inputs for automatic speech recognition without manually writing the default transcription prompt.
 
@@ -195,12 +196,18 @@ def apply_transcription_request(
             prompt (`str` or `list[str]`, *optional*):
                 Custom prompt(s) to include in the user turn. A list must be the same length as the batch. When `None`,
                 each sample uses `"Transcribe the input speech."`.
+            tokenize (`bool`, *optional*, defaults to `True`):
+                Whether to tokenize the prompts and process the audio. If `False`, the audio is not loaded and
+                the prompts are returned as strings with unexpanded audio placeholders. Passing them with the
+                audio to the processor, `processor(text=prompts, audio=audio)`, gives the same outputs as
+                `tokenize=True`.
             **kwargs:
                 Additional keyword arguments forwarded to [`~AudioFlamingo3Processor.apply_chat_template`] (for example
                 `text_kwargs`, `audio_kwargs`, ...).
 
         Returns:
-            [`BatchFeature`]: Processor outputs ready to be passed to [`AudioFlamingo3ForConditionalGeneration.generate`].
+            [`BatchFeature`] or `list[str]`: One prompt per audio when `tokenize=False`, otherwise processor
+            outputs ready to be passed to [`AudioFlamingo3ForConditionalGeneration.generate`].
 
         """
 
@@ -246,7 +253,7 @@ def apply_transcription_request(
 
         return self.apply_chat_template(
             conversations,
-            tokenize=True,
+            tokenize=tokenize,
             add_generation_prompt=True,
             return_dict=True,
             **kwargs,

src/transformers/models/canary/processing_canary.py

@@ -114,8 +114,9 @@ def apply_transcription_request(
         source_language: str | list[str] = "en",
         target_language: str | list[str] | None = None,
         punctuation: bool = True,
+        tokenize: bool = True,
         **kwargs: Unpack[CanaryProcessorKwargs],
-    ) -> BatchFeature:
+    ) -> BatchFeature | list[str]:
         r"""
         Prepare inputs for transcription or translation without manually writing the chat template.
 
@@ -130,11 +131,17 @@ def apply_transcription_request(
                 (transcription); set it to a different language for speech-to-text translation.
             punctuation (`bool`, *optional*, defaults to `True`):
                 Whether to request punctuation and capitalization in the output.
+            tokenize (`bool`, *optional*, defaults to `True`):
+                Whether to tokenize the prompts and process the audio. If `False`, the audio is not loaded and
+                the prompts are returned as strings with unexpanded audio placeholders. Passing them with the
+                audio to the processor, `processor(text=prompts, audio=audio)`, gives the same outputs as
+                `tokenize=True`.
             **kwargs:
                 Additional keyword arguments forwarded to [`~CanaryProcessor.apply_chat_template`].
 
         Returns:
-            [`BatchFeature`]: Processor outputs ready to be passed to
+            [`BatchFeature`] or `list[str]`: One prompt per audio when `tokenize=False`, otherwise processor
+            outputs ready to be passed to
             [`CanaryForConditionalGeneration.generate`].
         """
         audio_items = make_list_of_audio_chat_template(audio)
@@ -163,7 +170,7 @@ def apply_transcription_request(
 
         return self.apply_chat_template(
             conversations,
-            tokenize=True,
+            tokenize=tokenize,
             add_generation_prompt=True,
             return_dict=True,
             **kwargs,

src/transformers/models/fun_asr_nano/modular_fun_asr_nano.py

@@ -118,8 +118,9 @@ def apply_transcription_request(
         language: str | list[str] | None = None,
         prompt: str | list[str] | None = None,
         keywords: str | list[str] | list[list[str]] | None = None,
+        tokenize: bool = True,
         **kwargs: Unpack[FunAsrNanoProcessorKwargs],
-    ) -> BatchFeature:
+    ) -> BatchFeature | list[str]:
         """Prepare inputs for ASR using the checkpoint's structured transcription chat template.
 
         Args:
@@ -134,12 +135,18 @@ def apply_transcription_request(
             keywords (`str`, `list[str]`, or `list[list[str]]`, *optional*):
                 Hotwords to bias recognition. A string or flat list is shared across the batch; a nested list
                 supplies separate hotwords for each audio sample.
+            tokenize (`bool`, *optional*, defaults to `True`):
+                Whether to tokenize the prompts and process the audio. If `False`, the audio is not loaded and
+                the prompts are returned as strings with unexpanded audio placeholders. Passing them with the
+                audio to the processor, `processor(text=prompts, audio=audio)`, gives the same outputs as
+                `tokenize=True`.
             **kwargs:
                 Additional keyword arguments forwarded to [`~FunAsrNanoProcessor.apply_chat_template`]
                 and the underlying processor call (for example `text_kwargs`, `audio_kwargs`, ...).
 
         Returns:
-            [`BatchFeature`]: Processor outputs ready to be passed to
+            [`BatchFeature`] or `list[str]`: One prompt per audio when `tokenize=False`, otherwise processor
+            outputs ready to be passed to
             [`FunAsrNanoForConditionalGeneration.generate`].
         """
         audio_items = list(make_list_of_audio_chat_template(audio))
@@ -167,7 +174,7 @@ def apply_transcription_request(
 
         return self.apply_chat_template(
             conversations,
-            tokenize=True,
+            tokenize=tokenize,
             add_generation_prompt=True,
             return_dict=True,
             **kwargs,

src/transformers/models/fun_asr_nano/processing_fun_asr_nano.py

@@ -93,8 +93,9 @@ def apply_transcription_request(
         self,
         audio: str | list[str] | AudioInput,
         prompt: str | list[str] | None = None,
+        tokenize: bool = True,
         **kwargs: Unpack[GlmAsrProcessorKwargs],
-    ) -> BatchFeature:
+    ) -> BatchFeature | list[str]:
         """
         Prepare inputs for automatic speech recognition without manually writing the default transcription prompt.
 
@@ -105,12 +106,18 @@ def apply_transcription_request(
             prompt (`str` or `list[str]`, *optional*):
                 Custom prompt(s) to include in the user turn. A list must be the same length as the batch. When `None`,
                 each sample uses `"Transcribe the input speech."`.
+            tokenize (`bool`, *optional*, defaults to `True`):
+                Whether to tokenize the prompts and process the audio. If `False`, the audio is not loaded and
+                the prompts are returned as strings with unexpanded audio placeholders. Passing them with the
+                audio to the processor, `processor(text=prompts, audio=audio)`, gives the same outputs as
+                `tokenize=True`.
             **kwargs:
                 Additional keyword arguments forwarded to [`~GlmAsrProcessor.apply_chat_template`] (for example
                 `text_kwargs`, `audio_kwargs`, ...).
 
         Returns:
-            [`BatchFeature`]: Processor outputs ready to be passed to [`GlmAsrForConditionalGeneration.generate`].
+            [`BatchFeature`] or `list[str]`: One prompt per audio when `tokenize=False`, otherwise processor
+            outputs ready to be passed to [`GlmAsrForConditionalGeneration.generate`].
 
         """
 
@@ -156,7 +163,7 @@ def apply_transcription_request(
 
         return self.apply_chat_template(
             conversations,
-            tokenize=True,
+            tokenize=tokenize,
             add_generation_prompt=True,
             return_dict=True,
             **kwargs,

src/transformers/models/glmasr/modular_glmasr.py

@@ -442,8 +442,9 @@ def apply_transcription_request(
         audio: str | list[str] | AudioInput,
         prompt: str | list[str] | None = None,
         keywords: str | list[str] | list[list[str]] | None = None,
+        tokenize: bool = True,
         **kwargs: Unpack[MossTranscribeDiarizeProcessorKwargs],
-    ) -> BatchFeature:
+    ) -> BatchFeature | list[str]:
         """
         Prepare inputs for transcription / diarization without manually writing the chat template.
 
@@ -458,6 +459,11 @@ def apply_transcription_request(
                 Hotwords/domain terms (e.g. proper nouns) to bias transcription toward the correct spelling. A
                 string or flat list of strings is shared across the batch; a nested list supplies separate
                 keywords per audio sample.
+            tokenize (`bool`, *optional*, defaults to `True`):
+                Whether to tokenize the prompts and process the audio. If `False`, the audio is not loaded and
+                the prompts are returned as strings with unexpanded audio placeholders. Passing them with the
+                audio to the processor, `processor(text=prompts, audio=audio)`, gives the same outputs as
+                `tokenize=True`.
             **kwargs:
                 Additional keyword arguments forwarded to [`~MossTranscribeDiarizeProcessor.apply_chat_template`].
         """
@@ -482,7 +488,7 @@ def apply_transcription_request(
 
         return self.apply_chat_template(
             conversations,
-            tokenize=True,
+            tokenize=tokenize,
             add_generation_prompt=True,
             return_dict=True,
             **kwargs,
@@ -496,6 +502,9 @@ def parse_timestamp(value: str) -> float:
         """
         return sum(float(part) * 60**i for i, part in enumerate(reversed(value.strip().split(":"))))
 
+    def _decode_complete_segments(self, *args, **kwargs):
+        raise AttributeError("Not needed for MossTranscribeDiarize")
+
     def extract_speaker_dict(self, text: str | list[str]) -> list[dict] | str | list[list[dict] | str]:
         """
         Extract speaker/timestamp segments from raw output, returning original text on failure.

src/transformers/models/glmasr/processing_glmasr.py

@@ -442,8 +442,9 @@ def apply_transcription_request(
         audio: AudioInput | list[AudioInput],
         language: str | list[str] | None = None,
         prompt: str | list[str] | None = None,
+        tokenize: bool = True,
         **kwargs: Unpack[Qwen3ASRProcessorKwargs],
-    ) -> BatchFeature:
+    ) -> BatchFeature | list[str]:
         """
         Prepare inputs for automatic speech recognition without manually writing the chat template.
 
@@ -462,12 +463,18 @@ def apply_transcription_request(
                 Context/hotwords to include as the system prompt, e.g. domain-specific words or
                 phrases to bias the transcription towards. A list must be the same length as the
                 audio batch. When `None`, the system prompt is left empty.
+            tokenize (`bool`, *optional*, defaults to `True`):
+                Whether to tokenize the prompts and process the audio. If `False`, the audio is not loaded and
+                the prompts are returned as strings with unexpanded audio placeholders. Passing them with the
+                audio to the processor, `processor(text=prompts, audio=audio)`, gives the same outputs as
+                `tokenize=True`.
             **kwargs:
                 Additional keyword arguments forwarded to [`~Qwen3ASRProcessor.apply_chat_template`]
                 and the underlying processor call (for example `text_kwargs`, `audio_kwargs`, ...).
 
         Returns:
-            [`BatchFeature`]: Processor outputs ready to be passed to
+            [`BatchFeature`] or `list[str]`: One prompt per audio when `tokenize=False`, otherwise processor
+            outputs ready to be passed to
             [`Qwen3ASRForConditionalGeneration.generate`].
         """
         audio_items = list(make_list_of_audio_chat_template(audio))
@@ -494,7 +501,7 @@ def apply_transcription_request(
             messages.append({"role": "assistant", "content": [{"type": "text", "text": prefill}]})
         return self.apply_chat_template(
             conversations,
-            tokenize=True,
+            tokenize=tokenize,
             return_dict=True,
             continue_final_message=True,
             **kwargs,