Upload modeling_yor_audio.py with huggingface_hub

modeling_yor_audio.py

@@ -1,13 +1,63 @@
 import torch
 import torch.nn as nn
-from transformers import AutoModelForCausalLM, PreTrainedModel
+from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer, PreTrainedModel
 from transformers.models.qwen2_audio.modeling_qwen2_audio import Qwen2AudioEncoder
 from transformers.modeling_outputs import CausalLMOutputWithPast
-
-from configuration_yor_audio import YorAudioConfig
+from transformers import Qwen2AudioConfig, Qwen2AudioEncoderConfig
 
 AUDIO_TOKENS_PER_SEGMENT = 750
 
+AUDIO_SPECIAL_TOKENS = ["<|audio_bos|>", "<|audio_eos|>", "<|AUDIO|>"]
+
+WHISPER_LARGE_V3_AUDIO_CONFIG = {
+    "num_mel_bins": 128,
+    "encoder_layers": 32,
+    "encoder_attention_heads": 20,
+    "encoder_ffn_dim": 5120,
+    "d_model": 1280,
+    "dropout": 0.0,
+    "attention_dropout": 0.0,
+    "activation_function": "gelu",
+    "max_source_positions": 1500,
+    "initializer_range": 0.02,
+    "scale_embedding": False,
+}
+
+
+class YorAudioConfig(Qwen2AudioConfig):
+    model_type = "yor_audio"
+
+    def __init__(
+        self,
+        audio_config=None,
+        text_config=None,
+        audio_token_index: int = None,
+        afriqueqwen_path: str = "McGill-NLP/AfriqueQwen3.5-9B-50Langs-Instruct-v1",
+        **kwargs,
+    ):
+        if audio_config is None:
+            audio_config = Qwen2AudioEncoderConfig(**WHISPER_LARGE_V3_AUDIO_CONFIG)
+        elif isinstance(audio_config, dict):
+            audio_config = Qwen2AudioEncoderConfig(**audio_config)
+
+        if text_config is None:
+            from transformers import Qwen2Config
+            text_config = Qwen2Config()
+        elif isinstance(text_config, dict):
+            text_config = AutoConfig.for_model(**text_config)
+
+        if audio_token_index is None:
+            audio_token_index = text_config.vocab_size
+
+        self.afriqueqwen_path = afriqueqwen_path
+
+        super().__init__(
+            audio_config=audio_config,
+            text_config=text_config,
+            audio_token_index=audio_token_index,
+            **kwargs,
+        )
+
 
 class YorAudioMultiModalProjector(nn.Module):
     def __init__(self, config: YorAudioConfig):