transformers 5.0.0rc0__py3-none-any.whl → 5.0.0rc2__py3-none-any.whl

transformers/models/aria/modular_aria.py

@@ -1500,6 +1500,7 @@ class AriaForConditionalGeneration(LlavaForConditionalGeneration):
         attention_mask=None,
         cache_position=None,
         logits_to_keep=None,
+        is_first_iteration=False,
         **kwargs,
     ):
         model_inputs = super().prepare_inputs_for_generation(
@@ -1509,12 +1510,15 @@ class AriaForConditionalGeneration(LlavaForConditionalGeneration):
             attention_mask=attention_mask,
             cache_position=cache_position,
             logits_to_keep=logits_to_keep,
+            is_first_iteration=is_first_iteration,
             **kwargs,
         )
 
-        if cache_position[0] == 0:
-            # If we're in cached decoding stage, pixel values should be None because input ids do not contain special image token anymore
-            # Otherwise we need pixel values to be passed to model
+        if is_first_iteration or not kwargs.get("use_cache", True):
+            # Pixel values are used only in the first iteration if available
+            # In subsquent iterations, they are already merged with text and cached
+            # NOTE: first iteration doesn't have to be prefill, it can be the first
+            # iteration with a question and cached system prompt (continue generate from cache)
             model_inputs["pixel_values"] = pixel_values
             model_inputs["pixel_mask"] = pixel_mask
 

transformers/models/audioflamingo3/modeling_audioflamingo3.py

@@ -323,6 +323,7 @@ class AudioFlamingo3Encoder(AudioFlamingo3PreTrainedModel):
         self,
         input_features: torch.Tensor,
         input_features_mask: Optional[torch.Tensor] = None,
+        **kwargs,
     ):
         r"""
         Args:

transformers/models/audioflamingo3/modular_audioflamingo3.py

@@ -60,6 +60,7 @@ class AudioFlamingo3Encoder(Qwen2AudioEncoder):
         self,
         input_features: torch.Tensor,
         input_features_mask: Optional[torch.Tensor] = None,
+        **kwargs,
     ):
         r"""
         Args:

transformers/models/audioflamingo3/processing_audioflamingo3.py

@@ -32,9 +32,6 @@ if is_torch_available():
 
 logger = logging.get_logger(__name__)
 
-MAX_AUDIO_LEN = 10 * 60  # 10 minutes
-DEFAULT_TRANSCRIPTION_PROMPT = "Transcribe the input speech."
-
 
 class AudioFlamingo3ProcessorKwargs(ProcessingKwargs, total=False):
     _defaults = {
@@ -63,32 +60,41 @@ class AudioFlamingo3Processor(ProcessorMixin):
     [`Qwen2TokenizerFast`]. See the [`~AudioFlamingo3Processor.__call__`] for more information.
 
     Args:
-        feature_extractor ([`WhisperFeatureExtractor`]):
-            The feature extractor is a required input.
-        tokenizer ([`Qwen2TokenizerFast`]):
-            The tokenizer is a required input.
-        chat_template (`Optional[str]`, *optional*):
-            The Jinja template to use for formatting the conversation. If not provided, the tokenizer's default chat
-            template will be used.
-        audio_token (`Optional[str]`, *optional*, defaults to `"<sound>"`):
-            Special token used to represent audio inputs in the chat template.
+            feature_extractor ([`WhisperFeatureExtractor`]):
+                The feature extractor is a required input.
+            tokenizer ([`Qwen2TokenizerFast`]):
+                The tokenizer is a required input.
+            chat_template (`Optional[str]`, *optional*):
+                The Jinja template to use for formatting the conversation. If not provided, the tokenizer's default chat
+                template will be used.
+            audio_token (`Optional[str]`, *optional*, defaults to `"<sound>"`):
+                Special token used to represent audio inputs in the chat template.
+            default_transcription_prompt (`str`, *optional*, defaults to `"Transcribe the input speech."`):
+                Default prompt to use for transcription tasks when applying transcription requests.
+            max_audio_len (`int`, *optional*, defaults to 600):
+                Maximum length of audio sequences in seconds. Audio longer than this will be truncated.
     """
 
-    attributes = ["feature_extractor", "tokenizer"]
-    feature_extractor_class = "WhisperFeatureExtractor"
-    tokenizer_class = "Qwen2TokenizerFast"
-
     def __init__(
         self,
         feature_extractor,
         tokenizer,
         chat_template=None,
         audio_token="<sound>",
+        default_transcription_prompt="Transcribe the input speech.",
+        max_audio_len=600,
     ):
         self.audio_token = audio_token
         self.audio_token_id = tokenizer.convert_tokens_to_ids(audio_token)
+        self.default_transcription_prompt = default_transcription_prompt
+        self.max_audio_len = max_audio_len
         super().__init__(feature_extractor, tokenizer, chat_template=chat_template)
 
+    def _get_audio_token_length(self, audio_lengths: "torch.Tensor") -> "torch.Tensor":
+        conv_output_lengths = (audio_lengths - 1) // 2 + 1  # After conv2 downsampling
+        audio_tokens_lengths = (conv_output_lengths - 2) // 2 + 1  # After avg pooling
+        return audio_tokens_lengths
+
     def __call__(
         self,
         text: Union[TextInput, list[TextInput]],
@@ -143,7 +149,7 @@ class AudioFlamingo3Processor(ProcessorMixin):
 
             # Determine number of chunks per sample, and flatten
             window_size = int(audio_kwargs["sampling_rate"] * audio_kwargs["chunk_length"])
-            max_windows = int(MAX_AUDIO_LEN // audio_kwargs["chunk_length"])
+            max_windows = int(self.max_audio_len // audio_kwargs["chunk_length"])
 
             per_sample_windows: list[int] = []
             flat_chunks: list[np.ndarray] = []
@@ -153,7 +159,7 @@ class AudioFlamingo3Processor(ProcessorMixin):
                 n_win = max(1, (n_samples + window_size - 1) // window_size)
                 if n_win > max_windows:
                     logger.warning(
-                        f"Audio duration ({n_samples / audio_kwargs['sampling_rate']:.1f}s) exceeds {MAX_AUDIO_LEN}s; truncating to first {MAX_AUDIO_LEN}s."
+                        f"Audio duration ({n_samples / audio_kwargs['sampling_rate']:.1f}s) exceeds {self.max_audio_len}s; truncating to first {self.max_audio_len}s."
                     )
                     n_win = max_windows
                 per_sample_windows.append(n_win)
@@ -171,8 +177,7 @@ class AudioFlamingo3Processor(ProcessorMixin):
 
             # Compute sequence lengths token counting
             audio_lengths = torch.stack([s.sum() for s in torch.split(padding_mask.sum(-1), per_sample_windows)])
-            conv_output_lengths = (audio_lengths - 1) // 2 + 1  # After conv2 downsampling
-            audio_tokens_lengths = (conv_output_lengths - 2) // 2 + 1  # After avg pooling
+            audio_tokens_lengths = self._get_audio_token_length(audio_lengths)
 
             # expand audio tokens in text
             for i, audio_length in enumerate(audio_tokens_lengths):
@@ -236,7 +241,7 @@ class AudioFlamingo3Processor(ProcessorMixin):
             raise ValueError("`audio` must contain at least one sample.")
 
         if prompt is None:
-            prompts = [DEFAULT_TRANSCRIPTION_PROMPT] * batch_size
+            prompts = [self.default_transcription_prompt] * batch_size
         elif isinstance(prompt, str):
             prompts = [prompt] * batch_size
         elif isinstance(prompt, (list, tuple)):
@@ -247,7 +252,7 @@ class AudioFlamingo3Processor(ProcessorMixin):
             prompts = []
             for item in prompt:
                 if item is None:
-                    prompts.append(DEFAULT_TRANSCRIPTION_PROMPT)
+                    prompts.append(self.default_transcription_prompt)
                 elif isinstance(item, str):
                     prompts.append(item)
                 else:

transformers/models/auto/auto_factory.py

@@ -543,7 +543,7 @@ def add_generation_mixin_to_remote_model(model_class):
 
 class _LazyAutoMapping(OrderedDict[type[PreTrainedConfig], _LazyAutoMappingValue]):
     """
-    " A mapping config to object (model or tokenizer for instance) that will load keys and values when it is accessed.
+    A mapping config to object (model or tokenizer for instance) that will load keys and values when it is accessed.
 
     Args:
         - config_mapping: The map model type to config class

transformers/models/auto/configuration_auto.py

@@ -142,12 +142,14 @@ CONFIG_MAPPING_NAMES = OrderedDict[str, str](
         ("ernie", "ErnieConfig"),
         ("ernie4_5", "Ernie4_5Config"),
         ("ernie4_5_moe", "Ernie4_5_MoeConfig"),
+        ("ernie4_5_vl_moe", "Ernie4_5_VL_MoeConfig"),
         ("esm", "EsmConfig"),
         ("evolla", "EvollaConfig"),
         ("exaone4", "Exaone4Config"),
         ("falcon", "FalconConfig"),
         ("falcon_h1", "FalconH1Config"),
         ("falcon_mamba", "FalconMambaConfig"),
+        ("fast_vlm", "FastVlmConfig"),
         ("fastspeech2_conformer", "FastSpeech2ConformerConfig"),
         ("fastspeech2_conformer_with_hifigan", "FastSpeech2ConformerWithHifiGanConfig"),
         ("flaubert", "FlaubertConfig"),
@@ -178,6 +180,8 @@ CONFIG_MAPPING_NAMES = OrderedDict[str, str](
         ("glm4v_moe_vision", "Glm4vMoeVisionConfig"),
         ("glm4v_text", "Glm4vTextConfig"),
         ("glm4v_vision", "Glm4vVisionConfig"),
+        ("glmasr", "GlmAsrConfig"),
+        ("glmasr_encoder", "GlmAsrEncoderConfig"),
         ("glpn", "GLPNConfig"),
         ("got_ocr2", "GotOcr2Config"),
         ("gpt-sw3", "GPT2Config"),
@@ -214,12 +218,15 @@ CONFIG_MAPPING_NAMES = OrderedDict[str, str](
         ("instructblipvideo", "InstructBlipVideoConfig"),
         ("internvl", "InternVLConfig"),
         ("internvl_vision", "InternVLVisionConfig"),
+        ("jais2", "Jais2Config"),
         ("jamba", "JambaConfig"),
         ("janus", "JanusConfig"),
         ("jetmoe", "JetMoeConfig"),
         ("kosmos-2", "Kosmos2Config"),
         ("kosmos-2.5", "Kosmos2_5Config"),
         ("kyutai_speech_to_text", "KyutaiSpeechToTextConfig"),
+        ("lasr_ctc", "LasrCTCConfig"),
+        ("lasr_encoder", "LasrEncoderConfig"),
         ("layoutlm", "LayoutLMConfig"),
         ("layoutlmv2", "LayoutLMv2Config"),
         ("layoutlmv3", "LayoutLMv3Config"),
@@ -297,11 +304,18 @@ CONFIG_MAPPING_NAMES = OrderedDict[str, str](
         ("ovis2", "Ovis2Config"),
         ("owlv2", "Owlv2Config"),
         ("owlvit", "OwlViTConfig"),
+        ("paddleocr_vl", "PaddleOCRVLConfig"),
         ("paligemma", "PaliGemmaConfig"),
         ("parakeet_ctc", "ParakeetCTCConfig"),
         ("parakeet_encoder", "ParakeetEncoderConfig"),
         ("patchtsmixer", "PatchTSMixerConfig"),
         ("patchtst", "PatchTSTConfig"),
+        ("pe_audio", "PeAudioConfig"),
+        ("pe_audio_encoder", "PeAudioEncoderConfig"),
+        ("pe_audio_video", "PeAudioVideoConfig"),
+        ("pe_audio_video_encoder", "PeAudioVideoEncoderConfig"),
+        ("pe_video", "PeVideoConfig"),
+        ("pe_video_encoder", "PeVideoEncoderConfig"),
         ("pegasus", "PegasusConfig"),
         ("pegasus_x", "PegasusXConfig"),
         ("perceiver", "PerceiverConfig"),
@@ -312,6 +326,7 @@ CONFIG_MAPPING_NAMES = OrderedDict[str, str](
         ("phi4_multimodal", "Phi4MultimodalConfig"),
         ("phimoe", "PhimoeConfig"),
         ("pix2struct", "Pix2StructConfig"),
+        ("pixio", "PixioConfig"),
         ("pixtral", "PixtralVisionConfig"),
         ("plbart", "PLBartConfig"),
         ("poolformer", "PoolFormerConfig"),
@@ -578,6 +593,7 @@ MODEL_NAMES_MAPPING = OrderedDict[str, str](
         ("ernie", "ERNIE"),
         ("ernie4_5", "Ernie4_5"),
         ("ernie4_5_moe", "Ernie4_5_MoE"),
+        ("ernie4_5_vl_moe", "Ernie4_5_VL_MoE"),
         ("esm", "ESM"),
         ("evolla", "Evolla"),
         ("exaone4", "EXAONE-4.0"),
@@ -585,6 +601,7 @@ MODEL_NAMES_MAPPING = OrderedDict[str, str](
         ("falcon3", "Falcon3"),
         ("falcon_h1", "FalconH1"),
         ("falcon_mamba", "FalconMamba"),
+        ("fast_vlm", "FastVlm"),
         ("fastspeech2_conformer", "FastSpeech2Conformer"),
         ("fastspeech2_conformer_with_hifigan", "FastSpeech2ConformerWithHifiGan"),
         ("flan-t5", "FLAN-T5"),
@@ -617,6 +634,8 @@ MODEL_NAMES_MAPPING = OrderedDict[str, str](
         ("glm4v_moe_vision", "Glm4vMoeVisionModel"),
         ("glm4v_text", "GLM4V"),
         ("glm4v_vision", "Glm4vVisionModel"),
+        ("glmasr", "GLM-ASR"),
+        ("glmasr_encoder", "GLM-ASR Encoder"),
         ("glpn", "GLPN"),
         ("got_ocr2", "GOT-OCR2"),
         ("gpt-sw3", "GPT-Sw3"),
@@ -654,12 +673,16 @@ MODEL_NAMES_MAPPING = OrderedDict[str, str](
         ("instructblipvideo", "InstructBlipVideo"),
         ("internvl", "InternVL"),
         ("internvl_vision", "InternVLVision"),
+        ("jais2", "Jais2"),
         ("jamba", "Jamba"),
         ("janus", "Janus"),
         ("jetmoe", "JetMoe"),
         ("kosmos-2", "KOSMOS-2"),
         ("kosmos-2.5", "KOSMOS-2.5"),
         ("kyutai_speech_to_text", "KyutaiSpeechToText"),
+        ("lasr", "Lasr"),
+        ("lasr_ctc", "Lasr"),
+        ("lasr_encoder", "LasrEncoder"),
         ("layoutlm", "LayoutLM"),
         ("layoutlmv2", "LayoutLMv2"),
         ("layoutlmv3", "LayoutLMv3"),
@@ -747,12 +770,19 @@ MODEL_NAMES_MAPPING = OrderedDict[str, str](
         ("ovis2", "Ovis2"),
         ("owlv2", "OWLv2"),
         ("owlvit", "OWL-ViT"),
+        ("paddleocr_vl", "PaddleOCRVL"),
         ("paligemma", "PaliGemma"),
         ("parakeet", "Parakeet"),
         ("parakeet_ctc", "Parakeet"),
         ("parakeet_encoder", "ParakeetEncoder"),
         ("patchtsmixer", "PatchTSMixer"),
         ("patchtst", "PatchTST"),
+        ("pe_audio", "PeAudio"),
+        ("pe_audio_encoder", "PeAudioEncoder"),
+        ("pe_audio_video", "PeAudioVideo"),
+        ("pe_audio_video_encoder", "PeAudioVideoEncoder"),
+        ("pe_video", "PeVideo"),
+        ("pe_video_encoder", "PeVideoEncoder"),
         ("pegasus", "Pegasus"),
         ("pegasus_x", "PEGASUS-X"),
         ("perceiver", "Perceiver"),
@@ -764,6 +794,7 @@ MODEL_NAMES_MAPPING = OrderedDict[str, str](
         ("phimoe", "Phimoe"),
         ("phobert", "PhoBERT"),
         ("pix2struct", "Pix2Struct"),
+        ("pixio", "Pixio"),
         ("pixtral", "Pixtral"),
         ("plbart", "PLBart"),
         ("poolformer", "PoolFormer"),
@@ -946,6 +977,7 @@ SPECIAL_MODEL_TYPE_TO_MODULE_NAME = OrderedDict[str, str](
         ("glm4v_moe_vision", "glm4v_moe"),
         ("glm4v_text", "glm4v"),
         ("glm4v_moe_text", "glm4v_moe"),
+        ("glmasr_encoder", "glmasr"),
         ("grounding-dino", "grounding_dino"),
         ("mm-grounding-dino", "mm_grounding_dino"),
         ("idefics3_vision", "idefics3"),
@@ -972,9 +1004,15 @@ SPECIAL_MODEL_TYPE_TO_MODULE_NAME = OrderedDict[str, str](
         ("llama4_text", "llama4"),
         ("blip_2_qformer", "blip_2"),
         ("fastspeech2_conformer_with_hifigan", "fastspeech2_conformer"),
+        ("perception_encoder", "perception_lm"),
+        ("pe_audio_encoder", "pe_audio"),
+        ("pe_video_encoder", "pe_video"),
+        ("pe_audio_video_encoder", "pe_audio_video"),
         ("video_llama_3_vision", "video_llama_3"),
         ("parakeet_encoder", "parakeet"),
         ("parakeet_ctc", "parakeet"),
+        ("lasr_encoder", "lasr"),
+        ("lasr_ctc", "lasr"),
         ("wav2vec2-bert", "wav2vec2_bert"),
     ]
 )

transformers/models/auto/feature_extraction_auto.py

@@ -38,6 +38,7 @@ logger = logging.get_logger(__name__)
 FEATURE_EXTRACTOR_MAPPING_NAMES = OrderedDict(
     [
         ("audio-spectrogram-transformer", "ASTFeatureExtractor"),
+        ("audioflamingo3", "WhisperFeatureExtractor"),
         ("clap", "ClapFeatureExtractor"),
         ("clvp", "ClvpFeatureExtractor"),
         ("csm", "EncodecFeatureExtractor"),
@@ -46,9 +47,12 @@ FEATURE_EXTRACTOR_MAPPING_NAMES = OrderedDict(
         ("dia", "DiaFeatureExtractor"),
         ("encodec", "EncodecFeatureExtractor"),
         ("gemma3n", "Gemma3nAudioFeatureExtractor"),
+        ("glmasr", "WhisperFeatureExtractor"),
         ("granite_speech", "GraniteSpeechFeatureExtractor"),
         ("hubert", "Wav2Vec2FeatureExtractor"),
         ("kyutai_speech_to_text", "KyutaiSpeechToTextFeatureExtractor"),
+        ("lasr_ctc", "LasrFeatureExtractor"),
+        ("lasr_encoder", "LasrFeatureExtractor"),
         ("markuplm", "MarkupLMFeatureExtractor"),
         ("mimi", "EncodecFeatureExtractor"),
         ("moonshine", "Wav2Vec2FeatureExtractor"),
@@ -57,6 +61,8 @@ FEATURE_EXTRACTOR_MAPPING_NAMES = OrderedDict(
         ("musicgen_melody", "MusicgenMelodyFeatureExtractor"),
         ("parakeet_ctc", "ParakeetFeatureExtractor"),
         ("parakeet_encoder", "ParakeetFeatureExtractor"),
+        ("pe_audio", "PeAudioFeatureExtractor"),
+        ("pe_audio_video", "PeAudioFeatureExtractor"),
         ("phi4_multimodal", "Phi4MultimodalFeatureExtractor"),
         ("pop2piano", "Pop2PianoFeatureExtractor"),
         ("qwen2_5_omni", "WhisperFeatureExtractor"),
@@ -343,13 +349,13 @@ class AutoFeatureExtractor:
             )
             _ = kwargs.pop("code_revision", None)
             feature_extractor_class.register_for_auto_class()
-            return feature_extractor_class.from_dict(config_dict, **kwargs)
+            return feature_extractor_class.from_pretrained(pretrained_model_name_or_path, **kwargs)
         elif feature_extractor_class is not None:
-            return feature_extractor_class.from_dict(config_dict, **kwargs)
+            return feature_extractor_class.from_pretrained(pretrained_model_name_or_path, **kwargs)
         # Last try: we use the FEATURE_EXTRACTOR_MAPPING.
         elif type(config) in FEATURE_EXTRACTOR_MAPPING:
             feature_extractor_class = FEATURE_EXTRACTOR_MAPPING[type(config)]
-            return feature_extractor_class.from_dict(config_dict, **kwargs)
+            return feature_extractor_class.from_pretrained(pretrained_model_name_or_path, **kwargs)
 
         raise ValueError(
             f"Unrecognized feature extractor in {pretrained_model_name_or_path}. Should have a "

transformers/models/auto/image_processing_auto.py

@@ -100,6 +100,7 @@ else:
             ("efficientnet", ("EfficientNetImageProcessor", "EfficientNetImageProcessorFast")),
             ("emu3", ("Emu3ImageProcessor", None)),
             ("eomt", ("EomtImageProcessor", "EomtImageProcessorFast")),
+            ("ernie4_5_vl_moe", ("Ernie4_5_VL_MoeImageProcessor", "Ernie4_5_VL_MoeImageProcessorFast")),
             ("flava", ("FlavaImageProcessor", "FlavaImageProcessorFast")),
             ("florence2", ("CLIPImageProcessor", "CLIPImageProcessorFast")),
             ("focalnet", ("BitImageProcessor", "BitImageProcessorFast")),
@@ -153,11 +154,13 @@ else:
             ("ovis2", ("Ovis2ImageProcessor", "Ovis2ImageProcessorFast")),
             ("owlv2", ("Owlv2ImageProcessor", "Owlv2ImageProcessorFast")),
             ("owlvit", ("OwlViTImageProcessor", "OwlViTImageProcessorFast")),
+            ("paddleocr_vl", ("PaddleOCRVLImageProcessor", "PaddleOCRVLImageProcessorFast")),
             ("paligemma", ("SiglipImageProcessor", "SiglipImageProcessorFast")),
             ("perceiver", ("PerceiverImageProcessor", "PerceiverImageProcessorFast")),
             ("perception_lm", (None, "PerceptionLMImageProcessorFast")),
             ("phi4_multimodal", (None, "Phi4MultimodalImageProcessorFast")),
             ("pix2struct", ("Pix2StructImageProcessor", "Pix2StructImageProcessorFast")),
+            ("pixio", ("BitImageProcessor", "BitImageProcessorFast")),
             ("pixtral", ("PixtralImageProcessor", "PixtralImageProcessorFast")),
             ("poolformer", ("PoolFormerImageProcessor", "PoolFormerImageProcessorFast")),
             ("prompt_depth_anything", ("PromptDepthAnythingImageProcessor", "PromptDepthAnythingImageProcessorFast")),
@@ -604,9 +607,9 @@ class AutoImageProcessor:
             image_processor_class = get_class_from_dynamic_module(class_ref, pretrained_model_name_or_path, **kwargs)
             _ = kwargs.pop("code_revision", None)
             image_processor_class.register_for_auto_class()
-            return image_processor_class.from_dict(config_dict, **kwargs)
+            return image_processor_class.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
         elif image_processor_class is not None:
-            return image_processor_class.from_dict(config_dict, **kwargs)
+            return image_processor_class.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
         # Last try: we use the IMAGE_PROCESSOR_MAPPING.
         elif type(config) in IMAGE_PROCESSOR_MAPPING:
             image_processor_tuple = IMAGE_PROCESSOR_MAPPING[type(config)]

transformers/models/auto/modeling_auto.py

@@ -145,12 +145,14 @@ MODEL_MAPPING_NAMES = OrderedDict(
         ("ernie", "ErnieModel"),
         ("ernie4_5", "Ernie4_5Model"),
         ("ernie4_5_moe", "Ernie4_5_MoeModel"),
+        ("ernie4_5_vl_moe", "Ernie4_5_VL_MoeModel"),
         ("esm", "EsmModel"),
         ("evolla", "EvollaModel"),
         ("exaone4", "Exaone4Model"),
         ("falcon", "FalconModel"),
         ("falcon_h1", "FalconH1Model"),
         ("falcon_mamba", "FalconMambaModel"),
+        ("fast_vlm", "FastVlmModel"),
         ("fastspeech2_conformer", "FastSpeech2ConformerModel"),
         ("fastspeech2_conformer_with_hifigan", "FastSpeech2ConformerWithHifiGan"),
         ("flaubert", "FlaubertModel"),
@@ -181,6 +183,8 @@ MODEL_MAPPING_NAMES = OrderedDict(
         ("glm4v_moe_vision", "Glm4vMoeVisionModel"),
         ("glm4v_text", "Glm4vTextModel"),
         ("glm4v_vision", "Glm4vVisionModel"),
+        ("glmasr", "GlmAsrForConditionalGeneration"),
+        ("glmasr_encoder", "GlmAsrEncoder"),
         ("glpn", "GLPNModel"),
         ("got_ocr2", "GotOcr2Model"),
         ("gpt-sw3", "GPT2Model"),
@@ -215,12 +219,15 @@ MODEL_MAPPING_NAMES = OrderedDict(
         ("instructblipvideo", "InstructBlipVideoModel"),
         ("internvl", "InternVLModel"),
         ("internvl_vision", "InternVLVisionModel"),
+        ("jais2", "Jais2Model"),
         ("jamba", "JambaModel"),
         ("janus", "JanusModel"),
         ("jetmoe", "JetMoeModel"),
         ("kosmos-2", "Kosmos2Model"),
         ("kosmos-2.5", "Kosmos2_5Model"),
         ("kyutai_speech_to_text", "KyutaiSpeechToTextModel"),
+        ("lasr_ctc", "LasrForCTC"),
+        ("lasr_encoder", "LasrEncoder"),
         ("layoutlm", "LayoutLMModel"),
         ("layoutlmv2", "LayoutLMv2Model"),
         ("layoutlmv3", "LayoutLMv3Model"),
@@ -301,6 +308,12 @@ MODEL_MAPPING_NAMES = OrderedDict(
         ("parakeet_encoder", "ParakeetEncoder"),
         ("patchtsmixer", "PatchTSMixerModel"),
         ("patchtst", "PatchTSTModel"),
+        ("pe_audio", "PeAudioModel"),
+        ("pe_audio_encoder", "PeAudioEncoder"),
+        ("pe_audio_video", "PeAudioVideoModel"),
+        ("pe_audio_video_encoder", "PeAudioVideoEncoder"),
+        ("pe_video", "PeVideoModel"),
+        ("pe_video_encoder", "PeVideoEncoder"),
         ("pegasus", "PegasusModel"),
         ("pegasus_x", "PegasusXModel"),
         ("perceiver", "PerceiverModel"),
@@ -310,6 +323,7 @@ MODEL_MAPPING_NAMES = OrderedDict(
         ("phi3", "Phi3Model"),
         ("phi4_multimodal", "Phi4MultimodalModel"),
         ("phimoe", "PhimoeModel"),
+        ("pixio", "PixioModel"),
         ("pixtral", "PixtralVisionModel"),
         ("plbart", "PLBartModel"),
         ("poolformer", "PoolFormerModel"),
@@ -467,6 +481,7 @@ MODEL_FOR_PRETRAINING_MAPPING_NAMES = OrderedDict(
         ("fsmt", "FSMTForConditionalGeneration"),
         ("funnel", "FunnelForPreTraining"),
         ("gemma3", "Gemma3ForConditionalGeneration"),
+        ("glmasr", "GlmAsrForConditionalGeneration"),
         ("gpt-sw3", "GPT2LMHeadModel"),
         ("gpt2", "GPT2LMHeadModel"),
         ("gpt_bigcode", "GPTBigCodeForCausalLM"),
@@ -686,6 +701,7 @@ MODEL_FOR_CAUSAL_LM_MAPPING_NAMES = OrderedDict(
         ("helium", "HeliumForCausalLM"),
         ("hunyuan_v1_dense", "HunYuanDenseV1ForCausalLM"),
         ("hunyuan_v1_moe", "HunYuanMoEV1ForCausalLM"),
+        ("jais2", "Jais2ForCausalLM"),
         ("jamba", "JambaForCausalLM"),
         ("jetmoe", "JetMoeForCausalLM"),
         ("lfm2", "Lfm2ForCausalLM"),
@@ -795,6 +811,7 @@ MODEL_FOR_IMAGE_MAPPING_NAMES = OrderedDict(
         ("mobilenet_v2", "MobileNetV2Model"),
         ("mobilevit", "MobileViTModel"),
         ("mobilevitv2", "MobileViTV2Model"),
+        ("pixio", "PixioModel"),
         ("poolformer", "PoolFormerModel"),
         ("pvt", "PvtModel"),
         ("regnet", "RegNetModel"),
@@ -995,7 +1012,9 @@ MODEL_FOR_IMAGE_TEXT_TO_TEXT_MAPPING_NAMES = OrderedDict(
         ("deepseek_vl", "DeepseekVLForConditionalGeneration"),
         ("deepseek_vl_hybrid", "DeepseekVLHybridForConditionalGeneration"),
         ("emu3", "Emu3ForConditionalGeneration"),
+        ("ernie4_5_vl_moe", "Ernie4_5_VL_MoeForConditionalGeneration"),
         ("evolla", "EvollaForProteinText2Text"),
+        ("fast_vlm", "FastVlmForConditionalGeneration"),
         ("florence2", "Florence2ForConditionalGeneration"),
         ("fuyu", "FuyuForCausalLM"),
         ("gemma3", "Gemma3ForConditionalGeneration"),
@@ -1022,6 +1041,7 @@ MODEL_FOR_IMAGE_TEXT_TO_TEXT_MAPPING_NAMES = OrderedDict(
         ("mistral3", "Mistral3ForConditionalGeneration"),
         ("mllama", "MllamaForConditionalGeneration"),
         ("ovis2", "Ovis2ForConditionalGeneration"),
+        ("paddleocr_vl", "PaddleOCRVLForConditionalGeneration"),
         ("paligemma", "PaliGemmaForConditionalGeneration"),
         ("perception_lm", "PerceptionLMForConditionalGeneration"),
         ("pix2struct", "Pix2StructForConditionalGeneration"),
@@ -1045,6 +1065,7 @@ MODEL_FOR_IMAGE_TEXT_TO_TEXT_MAPPING_NAMES = OrderedDict(
 MODEL_FOR_MULTIMODAL_LM_MAPPING_NAMES = OrderedDict(
     [
         *list(MODEL_FOR_IMAGE_TEXT_TO_TEXT_MAPPING_NAMES.items()),
+        ("glmasr", "GlmAsrForConditionalGeneration"),
         ("granite_speech", "GraniteSpeechForConditionalGeneration"),
         ("kyutai_speech_to_text", "KyutaiSpeechToTextForConditionalGeneration"),
         ("phi4_multimodal", "Phi4MultimodalForCausalLM"),
@@ -1055,6 +1076,13 @@ MODEL_FOR_MULTIMODAL_LM_MAPPING_NAMES = OrderedDict(
     ]
 )
 
+MODEL_FOR_MASKED_LM_MAPPING_NAMES = OrderedDict(
+    [
+        # Model for Masked LM mapping
+        ("albert", "AlbertForMaskedLM"),
+    ]
+)
+
 MODEL_FOR_MASKED_LM_MAPPING_NAMES = OrderedDict(
     [
         # Model for Masked LM mapping
@@ -1151,6 +1179,7 @@ MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES = OrderedDict(
         ("blenderbot-small", "BlenderbotSmallForConditionalGeneration"),
         ("encoder-decoder", "EncoderDecoderModel"),
         ("fsmt", "FSMTForConditionalGeneration"),
+        ("glmasr", "GlmAsrForConditionalGeneration"),
         ("granite_speech", "GraniteSpeechForConditionalGeneration"),
         ("led", "LEDForConditionalGeneration"),
         ("longt5", "LongT5ForConditionalGeneration"),
@@ -1176,6 +1205,12 @@ MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES = OrderedDict(
     ]
 )
 
+MODEL_FOR_SPEECH_SEQ_2_SEQ_MAPPING_NAMES = OrderedDict(
+    [
+        ("dia", "DiaForConditionalGeneration"),
+    ]
+)
+
 MODEL_FOR_SPEECH_SEQ_2_SEQ_MAPPING_NAMES = OrderedDict(
     [
         ("dia", "DiaForConditionalGeneration"),
@@ -1581,6 +1616,7 @@ MODEL_FOR_CTC_MAPPING_NAMES = OrderedDict(
         # Model for Connectionist temporal classification (CTC) mapping
         ("data2vec-audio", "Data2VecAudioForCTC"),
         ("hubert", "HubertForCTC"),
+        ("lasr_ctc", "LasrForCTC"),
         ("parakeet_ctc", "ParakeetForCTC"),
         ("sew", "SEWForCTC"),
         ("sew-d", "SEWDForCTC"),
@@ -1674,6 +1710,7 @@ MODEL_FOR_BACKBONE_MAPPING_NAMES = OrderedDict(
         ("hgnet_v2", "HGNetV2Backbone"),
         ("hiera", "HieraBackbone"),
         ("maskformer-swin", "MaskFormerSwinBackbone"),
+        ("pixio", "PixioBackbone"),
         ("pvt_v2", "PvtV2Backbone"),
         ("resnet", "ResNetBackbone"),
         ("rt_detr_resnet", "RTDetrResNetBackbone"),

transformers/models/auto/processing_auto.py

@@ -68,6 +68,7 @@ PROCESSOR_MAPPING_NAMES = OrderedDict(
         ("dia", "DiaProcessor"),
         ("edgetam", "Sam2Processor"),
         ("emu3", "Emu3Processor"),
+        ("ernie4_5_vl_moe", "Ernie4_5_VL_MoeProcessor"),
         ("evolla", "EvollaProcessor"),
         ("flava", "FlavaProcessor"),
         ("florence2", "Florence2Processor"),
@@ -78,6 +79,7 @@ PROCESSOR_MAPPING_NAMES = OrderedDict(
         ("glm46v", "Glm46VProcessor"),
         ("glm4v", "Glm4vProcessor"),
         ("glm4v_moe", "Glm4vProcessor"),
+        ("glmasr", "GlmAsrProcessor"),
         ("got_ocr2", "GotOcr2Processor"),
         ("granite_speech", "GraniteSpeechProcessor"),
         ("grounding-dino", "GroundingDinoProcessor"),
@@ -93,6 +95,8 @@ PROCESSOR_MAPPING_NAMES = OrderedDict(
         ("kosmos-2", "Kosmos2Processor"),
         ("kosmos-2.5", "Kosmos2_5Processor"),
         ("kyutai_speech_to_text", "KyutaiSpeechToTextProcessor"),
+        ("lasr_ctc", "LasrProcessor"),
+        ("lasr_encoder", "LasrProcessor"),
         ("layoutlmv2", "LayoutLMv2Processor"),
         ("layoutlmv3", "LayoutLMv3Processor"),
         ("layoutxlm", "LayoutXLMProcessor"),
@@ -114,6 +118,7 @@ PROCESSOR_MAPPING_NAMES = OrderedDict(
         ("ovis2", "Ovis2Processor"),
         ("owlv2", "Owlv2Processor"),
         ("owlvit", "OwlViTProcessor"),
+        ("paddleocr_vl", "PaddleOCRVLProcessor"),
         ("paligemma", "PaliGemmaProcessor"),
         ("perception_lm", "PerceptionLMProcessor"),
         ("phi4_multimodal", "Phi4MultimodalProcessor"),
@@ -317,7 +322,6 @@ class AutoProcessor:
                     processor_class = config_dict.get("processor_class", None)
                     if "AutoProcessor" in config_dict.get("auto_map", {}):
                         processor_auto_map = config_dict["auto_map"]["AutoProcessor"]
-
             # Saved as feature extractor
             if preprocessor_config_file is None:
                 preprocessor_config_file = cached_file(
@@ -345,16 +349,24 @@ class AutoProcessor:
                     processor_auto_map = config_dict["auto_map"]["AutoProcessor"]
 
         if processor_class is None:
-            # Otherwise, load config, if it can be loaded.
-            if not isinstance(config, PreTrainedConfig):
-                config = AutoConfig.from_pretrained(
-                    pretrained_model_name_or_path, trust_remote_code=trust_remote_code, **kwargs
-                )
+            # Last resort: try loading the model config to get processor_class.
+            # This handles cases where processor info is only in config.json (not in any
+            # preprocessor/tokenizer config files). AutoConfig.from_pretrained may raise
+            # ValueError if the model_type is unrecognized or the config is invalid -
+            # we catch and ignore this to allow fallback to AutoTokenizer/AutoImageProcessor.
+            try:
+                if not isinstance(config, PreTrainedConfig):
+                    config = AutoConfig.from_pretrained(
+                        pretrained_model_name_or_path, trust_remote_code=trust_remote_code, **kwargs
+                    )
 
-            # And check if the config contains the processor class.
-            processor_class = getattr(config, "processor_class", None)
-            if hasattr(config, "auto_map") and "AutoProcessor" in config.auto_map:
-                processor_auto_map = config.auto_map["AutoProcessor"]
+                processor_class = getattr(config, "processor_class", None)
+                if hasattr(config, "auto_map") and "AutoProcessor" in config.auto_map:
+                    processor_auto_map = config.auto_map["AutoProcessor"]
+            except ValueError:
+                # Config loading failed (unrecognized model_type, invalid config, etc.)
+                # Continue to fallback logic below (AutoTokenizer, AutoImageProcessor, etc.)
+                pass
 
         if processor_class is not None:
             processor_class = processor_class_from_name(processor_class)