Lasr model (#42648)

* nit on dac!

* first commit

* fixes

* passing all API tests

* nit

* update

* test update

* make style

* make

* fixes

* fix

* test update

* return attention mask by default

* doc update

* make

* make

* fix

Author: eustlb

docs/source/en/_toctree.yml

@@ -901,6 +901,8 @@
         title: Hubert
       - local: model_doc/kyutai_speech_to_text
         title: Kyutai Speech-To-Text
+      - local: model_doc/lasr
+        title: LASR
       - local: model_doc/mimi
         title: Mimi
       - local: model_doc/mms

docs/source/en/model_doc/lasr.md

@@ -0,0 +1,108 @@
+<!--Copyright 2025 The HuggingFace Inc. team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+<div class="flex flex-wrap space-x-1">
+<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+<img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+</div>
+
+# LASR
+
+## Overview
+
+TODO
+
+## Usage
+
+### Basic usage
+
+<hfoptions id="usage">
+<hfoption id="Pipeline">
+
+```py
+from transformers import pipeline
+
+pipe = pipeline("automatic-speech-recognition", model="path/to/lasr-model")
+out = pipe("path/to/audio.mp3")
+print(out)
+```
+
+</hfoption>
+<hfoption id="AutoModel">
+
+```py
+from transformers import AutoModelForCTC, AutoProcessor
+from datasets import load_dataset, Audio
+import torch
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+processor = AutoProcessor.from_pretrained("path/to/lasr-model")
+model = AutoModelForCTC.from_pretrained("path/to/lasr-model", dtype="auto", device_map=device)
+
+ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
+ds = ds.cast_column("audio", Audio(sampling_rate=processor.feature_extractor.sampling_rate))
+speech_samples = [el['array'] for el in ds["audio"][:5]]
+
+inputs = processor(speech_samples, sampling_rate=processor.feature_extractor.sampling_rate)
+inputs.to(model.device, dtype=model.dtype)
+outputs = model.generate(**inputs)
+print(processor.batch_decode(outputs))
+```
+
+</hfoption>
+</hfoptions>
+
+### Making The Model Go Brrr
+
+TODO
+
+### Training
+
+TODO
+
+## LasrTokenizer
+
+[[autodoc]] LasrTokenizer
+
+## LasrFeatureExtractor
+
+[[autodoc]] LasrFeatureExtractor
+    - __call__
+
+## LasrProcessor
+
+[[autodoc]] LasrProcessor
+    - __call__
+    - batch_decode
+    - decode
+
+## LasrEncoderConfig
+
+[[autodoc]] LasrEncoderConfig
+
+## LasrCTCConfig
+
+[[autodoc]] LasrCTCConfig
+
+## LasrEncoder
+
+[[autodoc]] LasrEncoder
+
+## LasrForCTC
+
+[[autodoc]] LasrForCTC
+

src/transformers/models/__init__.py

@@ -186,6 +186,7 @@
     from .jetmoe import *
     from .kosmos2 import *
     from .kyutai_speech_to_text import *
+    from .lasr import *
     from .layoutlm import *
     from .layoutlmv2 import *
     from .layoutlmv3 import *

src/transformers/models/auto/configuration_auto.py

@@ -221,6 +221,8 @@
         ("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"),
@@ -662,6 +664,9 @@
         ("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"),
@@ -977,6 +982,8 @@
         ("video_llama_3_vision", "video_llama_3"),
         ("parakeet_encoder", "parakeet"),
         ("parakeet_ctc", "parakeet"),
+        ("lasr_encoder", "lasr"),
+        ("lasr_ctc", "lasr"),
         ("wav2vec2-bert", "wav2vec2_bert"),
     ]
 )

src/transformers/models/auto/feature_extraction_auto.py

@@ -49,6 +49,8 @@
         ("granite_speech", "GraniteSpeechFeatureExtractor"),
         ("hubert", "Wav2Vec2FeatureExtractor"),
         ("kyutai_speech_to_text", "KyutaiSpeechToTextFeatureExtractor"),
+        ("lasr_ctc", "LasrFeatureExtractor"),
+        ("lasr_encoder", "LasrFeatureExtractor"),
         ("markuplm", "MarkupLMFeatureExtractor"),
         ("mimi", "EncodecFeatureExtractor"),
         ("moonshine", "Wav2Vec2FeatureExtractor"),

src/transformers/models/auto/modeling_auto.py

@@ -222,6 +222,8 @@ class _BaseModelWithGenerate(PreTrainedModel, GenerationMixin):
         ("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"),
@@ -1583,6 +1585,7 @@ class _BaseModelWithGenerate(PreTrainedModel, GenerationMixin):
         # Model for Connectionist temporal classification (CTC) mapping
         ("data2vec-audio", "Data2VecAudioForCTC"),
         ("hubert", "HubertForCTC"),
+        ("lasr_ctc", "LasrForCTC"),
         ("parakeet_ctc", "ParakeetForCTC"),
         ("sew", "SEWForCTC"),
         ("sew-d", "SEWDForCTC"),

src/transformers/models/dac/modeling_dac.py

@@ -264,7 +264,7 @@ def forward(self, hidden_state):
         return hidden_state
 
 
-class DacResidualVectorQuantize(nn.Module):
+class DacResidualVectorQuantizer(nn.Module):
     """
     ResidualVectorQuantize block - Introduced in SoundStream: An end2end neural audio codec (https://huggingface.co/papers/2107.03312)
     """
@@ -568,7 +568,7 @@ def __init__(self, config: DacConfig):
         self.encoder = DacEncoder(config)
         self.decoder = DacDecoder(config)
 
-        self.quantizer = DacResidualVectorQuantize(config)
+        self.quantizer = DacResidualVectorQuantizer(config)
 
         self.bits_per_codebook = int(math.log2(self.config.codebook_size))
         if 2**self.bits_per_codebook != self.config.codebook_size:

src/transformers/models/fastspeech2_conformer/modeling_fastspeech2_conformer.py

@@ -514,7 +514,7 @@ def forward(self, hidden_states, attention_mask=None):
 
         Args:
             hidden_states (`torch.Tensor` of shape `(batch, time, channels)`): Input tensor.
-            attention_mask (`torch.Tensor` of shape `(batch, 1, time)`): Attention mask.
+            attention_mask (`torch.Tensor` of shape `(batch, 1, time, time)`): Attention mask.
 
         Returns:
             `torch.Tensor`: Output tensor of shape `(batch, time, channels)`.
@@ -530,7 +530,10 @@ def forward(self, hidden_states, attention_mask=None):
 
         # Apply padding mask before convolution
         if attention_mask is not None:
-            all_masked_rows = torch.all(~attention_mask, dim=-1)
+            if attention_mask.dtype == torch.bool:
+                all_masked_rows = torch.all(~attention_mask, dim=2)
+            else:
+                all_masked_rows = torch.all(~(attention_mask == 0.0), dim=2)
             hidden_states = hidden_states.masked_fill(all_masked_rows, 0.0)
 
         # 1D Depthwise Conv

src/transformers/models/lasr/__init__.py

@@ -0,0 +1,29 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from typing import TYPE_CHECKING
+
+from ...utils import _LazyModule
+from ...utils.import_utils import define_import_structure
+
+
+if TYPE_CHECKING:
+    from .configuration_lasr import *
+    from .feature_extraction_lasr import *
+    from .modeling_lasr import *
+    from .tokenization_lasr import *
+else:
+    import sys
+
+    _file = globals()["__file__"]
+    sys.modules[__name__] = _LazyModule(__name__, _file, define_import_structure(_file), module_spec=__spec__)