Upload folder using huggingface_hub

modeling_midashenglm.py

@@ -394,12 +394,10 @@ class Qwen25OmniTextModelOutput(ModelOutput):
 
 class Qwen25OmniThinkerTextOnlyDecoder(PreTrainedModel, GenerationMixin):
     config_class = Qwen2_5OmniTextConfig
-    _supports_flash_attn_2 = Qwen2_5OmniThinkerTextModel._supports_flash_attn_2
-    _supports_sdpa = Qwen2_5OmniThinkerTextModel._supports_sdpa
-    _supports_flex_attn = Qwen2_5OmniThinkerTextModel._supports_flex_attn
-    _supports_cache_class = Qwen2_5OmniThinkerTextModel._supports_cache_class
-    _supports_static_cache = Qwen2_5OmniThinkerTextModel._supports_static_cache
-    _supports_quantized_cache = Qwen2_5OmniThinkerTextModel._supports_quantized_cache
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
+    _supports_cache_class = True
+    _supports_static_cache = True
 
     def __init__(self, config: Qwen2_5OmniTextConfig):
         super().__init__(config)
@@ -471,15 +469,11 @@ class Qwen25OmniThinkerTextOnlyDecoder(PreTrainedModel, GenerationMixin):
 
 class MiDashengLMModel(PreTrainedModel):
     config_class = MiDashengLMConfig
-    _supports_flash_attn_2 = Qwen2_5OmniThinkerTextModel._supports_flash_attn_2
-    _supports_sdpa = Qwen2_5OmniThinkerTextModel._supports_sdpa
-    _supports_flex_attn = Qwen2_5OmniThinkerTextModel._supports_flex_attn
-    _supports_cache_class = Qwen2_5OmniThinkerTextModel._supports_cache_class
-    _supports_static_cache = Qwen2_5OmniThinkerTextModel._supports_static_cache
-    _supports_quantized_cache = Qwen2_5OmniThinkerTextModel._supports_quantized_cache
-    supports_gradient_checkpointing = (
-        Qwen2_5OmniThinkerTextModel.supports_gradient_checkpointing
-    )
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
+    _supports_cache_class = True
+    _supports_static_cache = True
+    supports_gradient_checkpointing = True
 
     def __init__(self, config: MiDashengLMConfig):
         super().__init__(config)
@@ -501,6 +495,12 @@ class MiDashengLMModel(PreTrainedModel):
 
         self.post_init()
 
+    def get_input_embeddings(self):
+        return self.decoder.model.embed_tokens
+
+    def get_output_embeddings(self):
+        return self.decoder.lm_head
+
     def _forward_audio_encoder(
         self,
         audios: torch.Tensor,