add kyutai_speech_to_text

Author: alien-0119

mindone/transformers/__init__.py

@@ -728,6 +728,11 @@
     JetMoePreTrainedModel,
 )
 from .models.kosmos2 import Kosmos2ForConditionalGeneration, Kosmos2Model, Kosmos2PreTrainedModel
+from .models.kyutai_speech_to_text import (
+    KyutaiSpeechToTextForConditionalGeneration,
+    KyutaiSpeechToTextModel,
+    KyutaiSpeechToTextPreTrainedModel,
+)
 from .models.layoutlm import (
     LayoutLMForMaskedLM,
     LayoutLMForQuestionAnswering,

mindone/transformers/generation/utils.py

@@ -2050,7 +2050,7 @@ def _supports_default_dynamic_cache(cls) -> bool:
         """
         # NOTE: remove xlnet/reformer when the models are deprecated, non-standard model architecture/cache name
         return not cls._is_stateful and all(
-            special_model_name not in cls.__name__.lower()
+            special_model_name not in cls.__class__.__name__.lower()
             for special_model_name in [
                 "reformer",
                 "minimax",

mindone/transformers/models/__init__.py

@@ -113,6 +113,7 @@
     jamba,
     jetmoe,
     kosmos2,
+    kyutai_speech_to_text,
     layoutlm,
     layoutlmv3,
     levit,

mindone/transformers/models/auto/configuration_auto.py

@@ -138,6 +138,7 @@
         ("jamba", "JambaConfig"),
         ("jetmoe", "JetMoeConfig"),
         ("kosmos-2", "Kosmos2Config"),
+        ("kyutai_speech_to_text", "KyutaiSpeechToTextConfig"),
         ("layoutlm", "LayoutLMConfig"),
         ("layoutlmv3", "LayoutLMv3Config"),
         ("led", "LEDConfig"),
@@ -406,6 +407,7 @@
         ("jamba", "Jamba"),
         ("jetmoe", "JetMoe"),
         ("kosmos-2", "KOSMOS-2"),
+        ("kyutai_speech_to_text", "KyutaiSpeechToText"),
         ("led", "LED"),
         ("levit", "LeViT"),
         ("lilt", "LiLT"),

mindone/transformers/models/auto/modeling_auto.py

@@ -130,6 +130,7 @@
         ("jamba", "JambaModel"),
         ("jetmoe", "JetMoeModel"),
         ("kosmos-2", "Kosmos2Model"),
+        ("kyutai_speech_to_text", "KyutaiSpeechToTextModel"),
         ("layoutlm", "LayoutLMModel"),
         ("layoutlmv3", "LayoutLMv3Model"),
         ("led", "LEDModel"),
@@ -838,6 +839,7 @@
 
 MODEL_FOR_SPEECH_SEQ_2_SEQ_MAPPING_NAMES = OrderedDict(
     [
+        ("kyutai_speech_to_text", "KyutaiSpeechToTextForConditionalGeneration"),
         ("moonshine", "MoonshineForConditionalGeneration"),
         ("pop2piano", "Pop2PianoForConditionalGeneration"),
         ("seamless_m4t", "SeamlessM4TForSpeechToText"),

mindone/transformers/models/kyutai_speech_to_text/__init__.py

@@ -0,0 +1,17 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# This code is adapted from https://github.com/huggingface/transformers
+# with modifications to run transformers on mindspore.
+#
+# 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 .modeling_kyutai_speech_to_text import *