Upload MOSS-Transcribe-Diarize pretrained model files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

added_tokens.json

@@ -0,0 +1,31 @@
+{
+  "</think>": 151668,
+  "</tool_call>": 151658,
+  "</tool_response>": 151666,
+  "<think>": 151667,
+  "<tool_call>": 151657,
+  "<tool_response>": 151665,
+  "<|audio_end|>": 151670,
+  "<|audio_pad|>": 151671,
+  "<|audio_start|>": 151669,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

chat_template.jinja

@@ -0,0 +1,103 @@
+{%- macro render_content(content) -%}
+    {%- if content is string -%}
+        {{- content -}}
+    {%- else -%}
+        {%- for item in content -%}
+            {%- if item.type == 'audio' or 'audio' in item or 'audio_url' in item -%}
+                {{- '<|audio_start|><|audio_pad|><|audio_end|>\n' -}}
+            {%- elif item.type == 'text' -%}
+                {{- item.text -}}
+            {%- endif -%}
+        {%- endfor -%}
+    {%- endif -%}
+{%- endmacro -%}
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0].role == 'system' %}
+        {{- render_content(messages[0].content) + '\n\n' }}
+    {%- else %}
+        {{- 'You are a helpful assistant.\n\n' }}
+    {%- endif %}
+    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {{- '<|im_start|>system\n' + render_content(messages[0].content) + '<|im_end|>\n' }}
+    {%- else %}
+        {{- '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- set content = render_content(message.content) %}
+    {%- if ns.multi_step_tool and message.role == "user" and content is string and not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+        {%- set ns.multi_step_tool = false %}
+        {%- set ns.last_query_index = index %}
+    {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content) %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {%- if loop.last or (not loop.last and reasoning_content) %}
+                {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
+            {%- else %}
+                {{- '<|im_start|>' + message.role + '\n' + content }}
+            {%- endif %}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if (loop.first and content) or (not loop.first) %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<tool_call>\n{"name": "' }}
+                {{- tool_call.name }}
+                {{- '", "arguments": ' }}
+                {%- if tool_call.arguments is string %}
+                    {{- tool_call.arguments }}
+                {%- else %}
+                    {{- tool_call.arguments | tojson }}
+                {%- endif %}
+                {{- '}\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}

config.json

@@ -0,0 +1,83 @@
+{
+  "architectures": [
+    "MossTranscribeDiarizeForConditionalGeneration"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_moss_transcribe_diarize.MossTranscribeDiarizeConfig",
+    "AutoModel": "modeling_moss_transcribe_diarize.MossTranscribeDiarizeModel",
+    "AutoModelForCausalLM": "modeling_moss_transcribe_diarize.MossTranscribeDiarizeForConditionalGeneration",
+    "AutoProcessor": "processing_moss_transcribe_diarize.MossTranscribeDiarizeProcessor"
+  },
+  "model_type": "moss_transcribe_diarize",
+  "text_config": {
+    "model_type": "qwen3",
+    "vocab_size": 151936,
+    "hidden_size": 1024,
+    "intermediate_size": 3072,
+    "num_hidden_layers": 28,
+    "num_attention_heads": 16,
+    "num_key_value_heads": 8,
+    "head_dim": 128,
+    "hidden_act": "silu",
+    "max_position_embeddings": 40960,
+    "initializer_range": 0.02,
+    "rms_norm_eps": 1e-06,
+    "tie_word_embeddings": true,
+    "rope_theta": 1000000,
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "layer_types": [
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention"
+    ],
+    "pad_token_id": 151643,
+    "bos_token_id": null,
+    "eos_token_id": null
+  },
+  "audio_config": {
+    "model_type": "whisper",
+    "num_mel_bins": 80,
+    "d_model": 1024,
+    "encoder_layers": 24,
+    "encoder_attention_heads": 16,
+    "encoder_ffn_dim": 4096,
+    "max_source_positions": 1500,
+    "dropout": 0.0,
+    "attention_dropout": 0.0,
+    "activation_dropout": 0.0,
+    "activation_function": "gelu",
+    "encoder_layerdrop": 0.0,
+    "scale_embedding": false
+  },
+  "audio_token_id": 151671,
+  "audio_merge_size": 4,
+  "adaptor_input_dim": 4096,
+  "tie_word_embeddings": true,
+  "pad_token_id": 151643
+}

configuration_moss_transcribe_diarize.py

@@ -0,0 +1,67 @@
+from transformers import PretrainedConfig
+from transformers.models.qwen3.configuration_qwen3 import Qwen3Config
+from transformers.models.whisper.configuration_whisper import WhisperConfig
+
+
+class MossTranscribeDiarizeConfig(PretrainedConfig):
+    """Configuration for MOSS-Transcribe-Diarize: Qwen3 text backbone + Whisper audio encoder."""
+
+    model_type = "moss_transcribe_diarize"
+    sub_configs = {"text_config": Qwen3Config, "audio_config": WhisperConfig}
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        text_config=None,
+        audio_config=None,
+        audio_token_id: int = 151671,
+        audio_merge_size: int = 4,
+        adaptor_input_dim: int | None = None,
+        tie_word_embeddings: bool = True,
+        **kwargs,
+    ):
+        if text_config is None:
+            text_config = Qwen3Config(
+                vocab_size=151936,
+                hidden_size=1024,
+                intermediate_size=3072,
+                num_hidden_layers=28,
+                num_attention_heads=16,
+                num_key_value_heads=8,
+                head_dim=128,
+                max_position_embeddings=40960,
+                tie_word_embeddings=tie_word_embeddings,
+                rope_theta=1_000_000.0,
+                layer_types=["full_attention"] * 28,
+            )
+        elif isinstance(text_config, dict):
+            text_config = self.sub_configs["text_config"](**text_config)
+
+        if audio_config is None:
+            audio_config = WhisperConfig(
+                num_mel_bins=80,
+                d_model=1024,
+                encoder_layers=24,
+                encoder_attention_heads=16,
+                encoder_ffn_dim=4096,
+                max_source_positions=1500,
+                dropout=0.0,
+                attention_dropout=0.0,
+                activation_dropout=0.0,
+                activation_function="gelu",
+                encoder_layerdrop=0.0,
+                scale_embedding=False,
+            )
+        elif isinstance(audio_config, dict):
+            audio_config = self.sub_configs["audio_config"](**audio_config)
+
+        text_config.tie_word_embeddings = tie_word_embeddings
+        if not getattr(text_config, "layer_types", None):
+            text_config.layer_types = ["full_attention"] * text_config.num_hidden_layers
+
+        self.text_config = text_config
+        self.audio_config = audio_config
+        self.audio_token_id = audio_token_id
+        self.audio_merge_size = audio_merge_size
+        self.adaptor_input_dim = adaptor_input_dim or audio_config.d_model * audio_merge_size
+        super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs)

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "bos_token_id": 151643,
+  "eos_token_id": 151645,
+  "pad_token_id": 151643,
+  "max_new_tokens": 5120
+}

model-00000-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b591a5d809df3d7e7ea4221811ecc35a739253213fa22e9bb7122137e1969f8
+size 1817113576

model.safetensors.index.json

@@ -0,0 +1,690 @@
+{
+  "metadata": {
+    "total_size": 1817026560
+  },
+  "weight_map": {
+    "model.whisper_encoder.conv1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.conv1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.conv2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.conv2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.embed_positions.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.0.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.1.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.10.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.11.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.12.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.13.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.14.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.15.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.16.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.17.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.18.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.19.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.2.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.20.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.21.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.22.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.23.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.3.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.4.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.5.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.6.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.7.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.8.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.fc1.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.fc1.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.fc2.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.fc2.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.final_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.final_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.self_attn.out_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.self_attn.out_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.self_attn.q_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.self_attn.v_proj.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.self_attn_layer_norm.bias": "model-00000-of-00001.safetensors",
+    "model.whisper_encoder.layers.9.self_attn_layer_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.0.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.0.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.0.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.0.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.0.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.0.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.0.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.0.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.0.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.0.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.0.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.1.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.1.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.1.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.1.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.1.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.1.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.1.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.1.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.1.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.1.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.1.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.10.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.10.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.10.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.10.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.10.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.10.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.10.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.10.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.10.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.10.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.10.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.11.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.11.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.11.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.11.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.11.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.11.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.11.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.11.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.11.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.11.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.11.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.12.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.12.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.12.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.12.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.12.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.12.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.12.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.12.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.12.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.12.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.12.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.13.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.13.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.13.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.13.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.13.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.13.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.13.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.13.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.13.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.13.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.13.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.14.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.14.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.14.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.14.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.14.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.14.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.14.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.14.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.14.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.14.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.14.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.15.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.15.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.15.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.15.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.15.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.15.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.15.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.15.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.15.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.15.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.15.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.16.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.16.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.16.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.16.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.16.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.16.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.16.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.16.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.16.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.16.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.16.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.17.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.17.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.17.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.17.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.17.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.17.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.17.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.17.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.17.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.17.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.17.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.18.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.18.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.18.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.18.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.18.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.18.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.18.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.18.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.18.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.18.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.18.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.19.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.19.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.19.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.19.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.19.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.19.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.19.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.19.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.19.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.19.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.19.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.2.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.2.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.2.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.2.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.2.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.2.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.2.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.2.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.2.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.2.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.2.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.20.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.20.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.20.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.20.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.20.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.20.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.20.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.20.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.20.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.20.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.20.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.21.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.21.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.21.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.21.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.21.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.21.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.21.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.21.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.21.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.21.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.21.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.22.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.22.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.22.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.22.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.22.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.22.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.22.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.22.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.22.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.22.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.22.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.23.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.23.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.23.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.23.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.23.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.23.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.23.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.23.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.23.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.23.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.23.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.24.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.24.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.24.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.24.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.24.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.24.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.24.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.24.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.24.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.24.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.24.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.25.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.25.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.25.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.25.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.25.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.25.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.25.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.25.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.25.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.25.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.25.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.26.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.26.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.26.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.26.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.26.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.26.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.26.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.26.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.26.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.26.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.26.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.27.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.27.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.27.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.27.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.27.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.27.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.27.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.27.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.27.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.27.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.27.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.3.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.3.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.3.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.3.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.3.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.3.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.3.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.3.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.3.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.3.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.3.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.4.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.4.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.4.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.4.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.4.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.4.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.4.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.4.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.4.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.4.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.4.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.5.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.5.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.5.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.5.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.5.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.5.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.5.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.5.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.5.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.5.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.5.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.6.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.6.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.6.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.6.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.6.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.6.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.6.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.6.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.6.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.6.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.6.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.7.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.7.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.7.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.7.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.7.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.7.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.7.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.7.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.7.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.7.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.7.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.8.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.8.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.8.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.8.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.8.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.8.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.8.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.8.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.8.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.8.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.8.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.9.post_attention_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.9.mlp.gate_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.9.mlp.up_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.9.mlp.down_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.9.self_attn.k_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.9.self_attn.o_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.9.input_layernorm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.9.self_attn.q_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.9.self_attn.k_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.9.self_attn.v_proj.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.layers.9.self_attn.q_norm.weight": "model-00000-of-00001.safetensors",
+    "model.language_model.embed_tokens.weight": "model-00000-of-00001.safetensors",
+    "model.vq_adaptor.layers.0.bias": "model-00000-of-00001.safetensors",
+    "model.vq_adaptor.layers.0.weight": "model-00000-of-00001.safetensors",
+    "model.vq_adaptor.layers.2.bias": "model-00000-of-00001.safetensors",
+    "model.vq_adaptor.layers.2.weight": "model-00000-of-00001.safetensors",
+    "model.vq_adaptor.layers.3.bias": "model-00000-of-00001.safetensors",
+    "model.vq_adaptor.layers.3.weight": "model-00000-of-00001.safetensors"
+  }
+}

modeling_moss_transcribe_diarize.py

@@ -0,0 +1,368 @@
+"""MossTranscribeDiarizeForConditionalGeneration: Whisper-Medium + VQAdaptor + Qwen3-0.6B.
+
+Architecture:
+    log-mel input_features -> HF WhisperEncoder
+             -> 4x time merge  (B, T, 1024) -> (B, T/4, 4096)
+             -> VQAdaptor       (4096 -> 1024)
+             -> masked_scatter into text embeddings
+             -> Qwen3-0.6B decoder -> logits
+"""
+
+from __future__ import annotations
+
+from typing import Optional
+
+import torch
+from torch import nn
+from transformers import GenerationMixin, PreTrainedModel
+from transformers.modeling_outputs import CausalLMOutputWithPast
+from transformers.models.qwen3.modeling_qwen3 import Qwen3Model
+from transformers.models.whisper.modeling_whisper import WhisperEncoder
+from transformers.utils import torch_compilable_check
+
+from .configuration_moss_transcribe_diarize import MossTranscribeDiarizeConfig
+
+
+class VQAdaptor(nn.Module):
+    """Projects merged Whisper features to LM hidden dim.
+
+    ``Linear(in → hidden) → SiLU → Linear(hidden → hidden) → LayerNorm``
+    """
+
+    def __init__(self, input_dim: int, hidden_size: int, norm_eps: float = 1e-6):
+        super().__init__()
+        self.layers = nn.Sequential(
+            nn.Linear(input_dim, hidden_size, bias=True),
+            nn.SiLU(),
+            nn.Linear(hidden_size, hidden_size, bias=True),
+            nn.LayerNorm(hidden_size, eps=norm_eps, bias=True),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.layers(x)
+
+
+class MossTranscribeDiarizePreTrainedModel(PreTrainedModel):
+    config_class = MossTranscribeDiarizeConfig
+    base_model_prefix = "model"
+    input_modalities = ("audio", "text")
+    _no_split_modules = ["Qwen3DecoderLayer", "WhisperEncoderLayer"]
+    _skip_keys_device_placement = "past_key_values"
+    supports_gradient_checkpointing = True
+    _supports_sdpa = True
+    _supports_attention_backend = True
+
+
+class MossTranscribeDiarizeModel(MossTranscribeDiarizePreTrainedModel):
+    base_model_prefix = "model"
+
+    """Single-stream multimodal backbone: Whisper-Medium encoder + Qwen3-0.6B.
+
+    Audio features are injected into text embeddings via ``masked_scatter`` at
+    positions marked by ``audio_token_id`` in ``input_ids``.
+    """
+
+    def __init__(self, config: MossTranscribeDiarizeConfig):
+        super().__init__(config)
+
+        self.language_model: nn.Module = Qwen3Model(config.text_config)
+        self.whisper_encoder: nn.Module = WhisperEncoder(config.audio_config)
+        self.vq_adaptor: VQAdaptor = VQAdaptor(
+            input_dim=config.adaptor_input_dim,
+            hidden_size=config.text_config.hidden_size,
+            norm_eps=config.text_config.rms_norm_eps,
+        )
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.language_model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.language_model.embed_tokens = value
+
+    # ---- 4x time merge ---------------------------------------------------
+
+    def time_merge(self, features: torch.Tensor) -> torch.Tensor:
+        """``(B, T, D) -> (B, T//M, D*M)`` where M is ``audio_merge_size``."""
+        B, T, D = features.shape
+        merge_size = int(self.config.audio_merge_size)
+        T_trim = (T // merge_size) * merge_size
+        return features[:, :T_trim, :].reshape(B, T_trim // merge_size, D * merge_size)
+
+    # ---- audio feature extraction -----------------------------------------
+
+    def get_audio_features(
+        self,
+        input_features: torch.Tensor,
+        audio_feature_lengths: torch.LongTensor,
+        audio_chunk_mapping: Optional[torch.LongTensor] = None,
+    ) -> list[torch.Tensor]:
+        """Whisper encoder -> 4x time merge -> VQAdaptor.
+
+        Returns list of ``(1, N_tokens, hidden_size)`` tensors.
+        """
+        if input_features is None:
+            raise ValueError("input_features must be provided for audio feature extraction.")
+        if audio_feature_lengths is None:
+            raise ValueError("audio_feature_lengths must be provided with input_features.")
+
+        device = next(self.whisper_encoder.parameters()).device
+        encoder_dtype = next(self.whisper_encoder.parameters()).dtype
+        input_features = input_features.to(device=device, dtype=encoder_dtype)
+        audio_feature_lengths = audio_feature_lengths.to(device=device)
+        if audio_feature_lengths.numel() != input_features.shape[0]:
+            raise ValueError(
+                "audio_feature_lengths must contain one length per input_features chunk: "
+                f"got {audio_feature_lengths.numel()} lengths for {input_features.shape[0]} chunks."
+            )
+
+        whisper_features = self.whisper_encoder(input_features, return_dict=True).last_hidden_state
+
+        chunk_mapping = (
+            audio_chunk_mapping.to(device=device)
+            if audio_chunk_mapping is not None
+            else torch.zeros(input_features.shape[0], dtype=torch.long, device=device)
+        )
+        if chunk_mapping.numel() != input_features.shape[0]:
+            raise ValueError(
+                "audio_chunk_mapping must contain one sample index per input_features chunk: "
+                f"got {chunk_mapping.numel()} indices for {input_features.shape[0]} chunks."
+            )
+
+        num_audios = int(chunk_mapping.max().item()) + 1 if chunk_mapping.numel() else 0
+        per_audio_chunks = [[] for _ in range(num_audios)]
+        for chunk_idx, token_len in enumerate(audio_feature_lengths.tolist()):
+            sample_idx = int(chunk_mapping[chunk_idx].item())
+            per_audio_chunks[sample_idx].append(
+                whisper_features[chunk_idx : chunk_idx + 1, : int(token_len) * 4]
+            )
+
+        adapted = []
+        for parts in per_audio_chunks:
+            feat = torch.cat(parts, dim=1)
+            feat = feat.to(self.dtype)
+            merged = self.time_merge(feat)
+            adapted.append(self.vq_adaptor(merged))
+        return adapted
+
+    # ---- inject audio into text embeddings --------------------------------
+
+    def get_placeholder_mask(
+        self,
+        input_ids: Optional[torch.LongTensor],
+        inputs_embeds: torch.FloatTensor,
+        audio_features: torch.Tensor,
+    ) -> torch.BoolTensor:
+        """Return the expanded audio placeholder mask and validate feature count."""
+        if input_ids is None:
+            special_audio_mask = inputs_embeds == self.get_input_embeddings()(
+                torch.tensor(self.config.audio_token_id, dtype=torch.long, device=inputs_embeds.device)
+            )
+            special_audio_mask = special_audio_mask.all(-1)
+        else:
+            special_audio_mask = input_ids.to(device=inputs_embeds.device) == self.config.audio_token_id
+
+        if special_audio_mask.shape != inputs_embeds.shape[:2]:
+            raise ValueError(
+                "input_ids shape must match the first two dimensions of inputs_embeds: "
+                f"got {tuple(special_audio_mask.shape)} and {tuple(inputs_embeds.shape[:2])}."
+            )
+
+        n_audio_tokens = special_audio_mask.sum()
+        special_audio_mask = special_audio_mask.unsqueeze(-1).expand_as(inputs_embeds).to(inputs_embeds.device)
+        torch_compilable_check(
+            inputs_embeds[special_audio_mask].numel() == audio_features.numel(),
+            (
+                f"Audio features and audio tokens do not match: "
+                f"tokens: {n_audio_tokens}, features {audio_features.shape[0]}"
+            ),
+        )
+        return special_audio_mask
+
+    def inject_audio_features(
+        self,
+        input_ids,
+        inputs_embeds,
+        input_features,
+        audio_feature_lengths,
+        audio_chunk_mapping,
+    ):
+        """Replace audio placeholder positions with projected audio features."""
+        if input_features is None:
+            return inputs_embeds
+        audio_features = self.get_audio_features(
+            input_features=input_features,
+            audio_feature_lengths=audio_feature_lengths,
+            audio_chunk_mapping=audio_chunk_mapping,
+        )
+        audio_embeds = torch.cat([f.squeeze(0) for f in audio_features], dim=0)
+        audio_embeds = audio_embeds.to(inputs_embeds.device, inputs_embeds.dtype)
+        audio_mask = self.get_placeholder_mask(input_ids, inputs_embeds, audio_embeds)
+        return inputs_embeds.masked_scatter(audio_mask, audio_embeds)
+
+    # ---- forward ----------------------------------------------------------
+
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values=None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        input_features: Optional[torch.FloatTensor] = None,
+        audio_feature_lengths: Optional[torch.LongTensor] = None,
+        audio_chunk_mapping: Optional[torch.LongTensor] = None,
+        **kwargs,
+    ):
+        return_dict = True if return_dict is None else return_dict
+        if input_ids is None and inputs_embeds is None:
+            raise ValueError("You must specify one of input_ids or inputs_embeds.")
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You must specify only one of input_ids or inputs_embeds.")
+
+        if inputs_embeds is None:
+            inputs_embeds = self.get_input_embeddings()(input_ids)
+        inputs_embeds = self.inject_audio_features(
+            input_ids=input_ids,
+            inputs_embeds=inputs_embeds,
+            input_features=input_features,
+            audio_feature_lengths=audio_feature_lengths,
+            audio_chunk_mapping=audio_chunk_mapping,
+        )
+        if output_attentions is not None:
+            kwargs["output_attentions"] = output_attentions
+        if output_hidden_states is not None:
+            kwargs["output_hidden_states"] = output_hidden_states
+
+        outputs = self.language_model(
+            input_ids=None, attention_mask=attention_mask, position_ids=position_ids,
+            past_key_values=past_key_values, inputs_embeds=inputs_embeds,
+            use_cache=use_cache, **kwargs,
+        )
+        if not return_dict:
+            return outputs.to_tuple()
+        return outputs
+
+
+class MossTranscribeDiarizeForConditionalGeneration(MossTranscribeDiarizePreTrainedModel, GenerationMixin):
+    _tied_weights_keys = {"lm_head.weight": "model.language_model.embed_tokens.weight"}
+
+    def __init__(self, config: MossTranscribeDiarizeConfig):
+        super().__init__(config)
+        self.model = MossTranscribeDiarizeModel(config)
+        self.vocab_size = config.text_config.vocab_size
+        self.lm_head = nn.Linear(config.text_config.hidden_size, config.text_config.vocab_size, bias=False)
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.get_input_embeddings()
+
+    def set_input_embeddings(self, value):
+        self.model.set_input_embeddings(value)
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def get_audio_features(
+        self,
+        input_features: torch.Tensor,
+        audio_feature_lengths: torch.LongTensor,
+        audio_chunk_mapping: Optional[torch.LongTensor] = None,
+    ) -> list[torch.Tensor]:
+        return self.model.get_audio_features(
+            input_features=input_features,
+            audio_feature_lengths=audio_feature_lengths,
+            audio_chunk_mapping=audio_chunk_mapping,
+        )
+
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values=None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        input_features: Optional[torch.FloatTensor] = None,
+        audio_feature_lengths: Optional[torch.LongTensor] = None,
+        audio_chunk_mapping: Optional[torch.LongTensor] = None,
+        logits_to_keep: int | torch.Tensor = 0,
+        **kwargs,
+    ):
+        return_dict = True if return_dict is None else return_dict
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=True,
+            input_features=input_features,
+            audio_feature_lengths=audio_feature_lengths,
+            audio_chunk_mapping=audio_chunk_mapping,
+            **kwargs,
+        )
+
+        hidden_states = outputs.last_hidden_state
+        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(
+                logits=logits,
+                labels=labels,
+                vocab_size=self.config.text_config.vocab_size,
+                **kwargs,
+            )
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+        return CausalLMOutputWithPast(
+            loss=loss, logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    # ---- generation support -----------------------------------------------
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        past_key_values=None,
+        attention_mask=None,
+        inputs_embeds=None,
+        input_features=None,
+        audio_feature_lengths=None,
+        audio_chunk_mapping=None,
+        is_first_iteration=False,
+        use_cache=True,
+        **kwargs,
+    ):
+        model_inputs = super().prepare_inputs_for_generation(
+            input_ids,
+            past_key_values=past_key_values,
+            attention_mask=attention_mask, inputs_embeds=inputs_embeds,
+            is_first_iteration=is_first_iteration, use_cache=use_cache, **kwargs,
+        )
+        if input_features is not None and (is_first_iteration or not use_cache):
+            model_inputs["input_features"] = input_features
+            model_inputs["audio_feature_lengths"] = audio_feature_lengths
+            model_inputs["audio_chunk_mapping"] = audio_chunk_mapping
+        return model_inputs

preprocessor_config.json

@@ -0,0 +1,14 @@
+{
+  "chunk_length": 30,
+  "dither": 0.0,
+  "feature_extractor_type": "WhisperFeatureExtractor",
+  "feature_size": 80,
+  "hop_length": 160,
+  "n_fft": 400,
+  "n_samples": 480000,
+  "nb_max_frames": 3000,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "return_attention_mask": false,
+  "sampling_rate": 16000
+}

processing_moss_transcribe_diarize.py

@@ -0,0 +1,272 @@
+"""Processor for MOSS-Transcribe-Diarize audio-text inference."""
+
+from __future__ import annotations
+
+from typing import Optional, Union
+
+import numpy as np
+import torch
+from transformers.feature_extraction_utils import BatchFeature
+from transformers.processing_utils import ProcessingKwargs, ProcessorMixin, Unpack
+
+
+AUDIO_PAD_TOKEN = "<|audio_pad|>"
+AUDIO_START_TOKEN = "<|audio_start|>"
+AUDIO_END_TOKEN = "<|audio_end|>"
+WHISPER_ENCODER_STRIDE = 2
+
+
+class MossTranscribeDiarizeProcessorKwargs(ProcessingKwargs, total=False):
+    _defaults = {
+        "common_kwargs": {
+            "return_tensors": "pt",
+        },
+        "audio_kwargs": {},
+    }
+
+
+def _audio_to_numpy(audio: Union[np.ndarray, torch.Tensor]) -> np.ndarray:
+    if torch.is_tensor(audio):
+        audio = audio.detach().cpu().numpy()
+    audio = np.asarray(audio, dtype=np.float32)
+    if audio.ndim > 1:
+        audio = np.squeeze(audio)
+    if audio.ndim == 0:
+        audio = audio.reshape(1)
+    if audio.ndim != 1:
+        raise ValueError(f"Expected mono audio with shape (num_samples,), got shape {audio.shape}.")
+    if audio.shape[0] == 0:
+        raise ValueError("Audio must contain at least one sample.")
+    return audio
+
+
+def _pad_or_trim_audio(audio: np.ndarray, length: int) -> np.ndarray:
+    if audio.shape[0] > length:
+        audio = audio[:length]
+    elif audio.shape[0] < length:
+        audio = np.pad(audio, (0, length - audio.shape[0]))
+    return audio.astype(np.float32, copy=False)
+
+
+def _compute_audio_token_length(num_samples: int, feature_extractor, audio_merge_size: int) -> int:
+    stride = int(feature_extractor.hop_length) * WHISPER_ENCODER_STRIDE * int(audio_merge_size)
+    return (int(num_samples) - 1) // stride + 1
+
+
+def _chunk_audio(
+    feature_extractor,
+    audio: Union[np.ndarray, torch.Tensor],
+    audio_merge_size: int,
+) -> tuple[np.ndarray, list[int]]:
+    audio = _audio_to_numpy(audio)
+    n_samples = int(feature_extractor.n_samples)
+
+    chunks, token_lengths = [], []
+    for start in range(0, audio.shape[0], n_samples):
+        chunk = audio[start : start + n_samples]
+        token_lengths.append(_compute_audio_token_length(chunk.shape[0], feature_extractor, audio_merge_size))
+        chunks.append(_pad_or_trim_audio(chunk, n_samples))
+    return np.stack(chunks), token_lengths
+
+
+def _audios_to_input_features(
+    feature_extractor,
+    audios: list[Union[np.ndarray, torch.Tensor]],
+    *,
+    audio_merge_size: int,
+    feature_extractor_kwargs: Optional[dict] = None,
+) -> tuple[torch.Tensor, torch.LongTensor, torch.LongTensor]:
+    feature_batches, feature_lengths, chunk_mapping = [], [], []
+    feature_extractor_kwargs = dict(feature_extractor_kwargs or {})
+    feature_extractor_kwargs.update(
+        {
+            "sampling_rate": int(feature_extractor.sampling_rate),
+            "padding": "max_length",
+            "return_tensors": "pt",
+        }
+    )
+
+    for audio_idx, audio in enumerate(audios):
+        chunks, token_lengths = _chunk_audio(feature_extractor, audio, audio_merge_size)
+        features = feature_extractor(
+            list(chunks),
+            **feature_extractor_kwargs,
+        )["input_features"]
+        feature_batches.append(features)
+        feature_lengths.extend(token_lengths)
+        chunk_mapping.extend([audio_idx] * len(token_lengths))
+
+    if feature_batches:
+        input_features = torch.cat(feature_batches, dim=0)
+    else:
+        input_features = torch.empty(
+            (0, int(feature_extractor.feature_size), int(feature_extractor.nb_max_frames)),
+        )
+
+    length_device = input_features.device
+    return (
+        input_features,
+        torch.tensor(feature_lengths, dtype=torch.long, device=length_device),
+        torch.tensor(chunk_mapping, dtype=torch.long, device=length_device),
+    )
+
+
+class MossTranscribeDiarizeProcessor(ProcessorMixin):
+    """Build MOSS-Transcribe-Diarize model inputs from text prompts and raw waveforms.
+
+    The model consumes log-mel ``input_features``. This processor owns the raw
+    waveform preprocessing, audio placeholder expansion, and optional numeric
+    time anchors inside the audio span.
+    """
+
+    attributes = ["feature_extractor", "tokenizer"]
+    feature_extractor_class = "AutoFeatureExtractor"
+    tokenizer_class = "AutoTokenizer"
+
+    model_input_names = [
+        "input_ids",
+        "attention_mask",
+        "input_features",
+        "audio_feature_lengths",
+        "audio_chunk_mapping",
+    ]
+
+    def __init__(
+        self,
+        feature_extractor=None,
+        tokenizer=None,
+        audio_tokens_per_second: float = 12.5,
+        audio_merge_size: int = 4,
+        time_marker_every_seconds: int = 2,
+        enable_time_marker: bool = True,
+        chat_template: Optional[str] = None,
+    ):
+        if feature_extractor is None:
+            raise ValueError("MossTranscribeDiarizeProcessor requires a feature_extractor.")
+        if tokenizer is None:
+            raise ValueError("MossTranscribeDiarizeProcessor requires a tokenizer.")
+        super().__init__(feature_extractor, tokenizer, chat_template=chat_template)
+        self.audio_tokens_per_second = audio_tokens_per_second
+        self.audio_merge_size = int(audio_merge_size)
+        self.time_marker_every_seconds = time_marker_every_seconds
+        self.enable_time_marker = enable_time_marker
+        self.audio_token = AUDIO_PAD_TOKEN if not hasattr(tokenizer, "audio_token") else tokenizer.audio_token
+        self.audio_start_token = (
+            AUDIO_START_TOKEN if not hasattr(tokenizer, "audio_start_token") else tokenizer.audio_start_token
+        )
+        self.audio_end_token = AUDIO_END_TOKEN if not hasattr(tokenizer, "audio_end_token") else tokenizer.audio_end_token
+        resolved_audio_token_id = tokenizer.convert_tokens_to_ids(self.audio_token)
+        if resolved_audio_token_id is None:
+            raise ValueError(f"Tokenizer is missing required audio placeholder token {self.audio_token!r}.")
+        self.audio_token_id = int(resolved_audio_token_id)
+        self.digit_token_ids = self._get_digit_token_ids()
+
+    def _get_digit_token_ids(self) -> dict[str, int]:
+        digit_token_ids = {}
+        for digit in "0123456789":
+            ids = self.tokenizer.encode(digit, add_special_tokens=False)
+            if len(ids) != 1:
+                raise ValueError(f"Digit {digit!r} is not a single token: {ids}")
+            digit_token_ids[digit] = int(ids[0])
+        return digit_token_ids
+
+    def _audio_span_ids(self, audio_seq_len: int) -> list[int]:
+        audio_seq_len = int(audio_seq_len)
+        if not self.enable_time_marker or audio_seq_len <= 0 or self.time_marker_every_seconds <= 0:
+            return [self.audio_token_id] * max(audio_seq_len, 0)
+
+        tokens_per_marker = int(self.audio_tokens_per_second * self.time_marker_every_seconds)
+        if tokens_per_marker <= 0:
+            return [self.audio_token_id] * audio_seq_len
+
+        duration = audio_seq_len / float(self.audio_tokens_per_second)
+        output, consumed = [], 0
+        for sec in range(self.time_marker_every_seconds, int(duration) + 1, self.time_marker_every_seconds):
+            pos = (sec // self.time_marker_every_seconds) * tokens_per_marker
+            segment_len = pos - consumed
+            if segment_len > 0:
+                output.extend([self.audio_token_id] * segment_len)
+                consumed += segment_len
+            marker_ids = [self.digit_token_ids[digit] for digit in str(sec)]
+            output.extend(marker_ids)
+
+        remainder = audio_seq_len - consumed
+        if remainder > 0:
+            output.extend([self.audio_token_id] * remainder)
+        return output
+
+    def _expand_audio_token(self, text: str, num_audio_tokens: int, max_length: int) -> list[int]:
+        audio_ids = self._audio_span_ids(num_audio_tokens)
+        audio_token_count = text.count(self.audio_token)
+        if audio_token_count != 1:
+            raise ValueError(
+                f"Expected exactly one {self.audio_token!r} token per text sample, got {audio_token_count}."
+            )
+        before_audio, after_audio = text.split(self.audio_token, maxsplit=1)
+        before_ids = self.tokenizer.encode(before_audio, add_special_tokens=False)
+        after_ids = self.tokenizer.encode(after_audio, add_special_tokens=False)
+
+        input_ids = before_ids + audio_ids + after_ids
+
+        if len(input_ids) > max_length:
+            raise ValueError(f"Prompt/audio sequence exceeds max_length={max_length}")
+        return input_ids
+
+    def __call__(
+        self,
+        text: Union[str, list[str]],
+        audio,
+        *,
+        max_length: int = 131072,
+        **kwargs: Unpack[MossTranscribeDiarizeProcessorKwargs],
+    ) -> BatchFeature:
+        return_tensors = kwargs.pop("return_tensors", "pt")
+        output_kwargs = self._merge_kwargs(
+            MossTranscribeDiarizeProcessorKwargs,
+            tokenizer_init_kwargs=self.tokenizer.init_kwargs,
+            **kwargs,
+        )
+        if isinstance(text, str):
+            texts = [text]
+        else:
+            texts = list(text)
+        audios = audio if isinstance(audio, list) else [audio]
+        if len(texts) != len(audios):
+            raise ValueError(f"Expected one audio per text prompt, got {len(audios)} audios and {len(texts)} prompts.")
+
+        input_features, audio_feature_lengths, audio_chunk_mapping = _audios_to_input_features(
+            self.feature_extractor,
+            audios,
+            audio_merge_size=self.audio_merge_size,
+            feature_extractor_kwargs=output_kwargs["audio_kwargs"],
+        )
+        audio_token_counts = torch.zeros(len(audios), dtype=torch.long, device=audio_feature_lengths.device)
+        audio_token_counts.scatter_add_(0, audio_chunk_mapping, audio_feature_lengths)
+
+        encoded = [
+            self._expand_audio_token(prompt, int(num_audio_tokens.item()), max_length)
+            for prompt, num_audio_tokens in zip(texts, audio_token_counts)
+        ]
+        max_seq_len = max(len(ids) for ids in encoded)
+        pad_token_id = self.tokenizer.pad_token_id
+        if pad_token_id is None:
+            pad_token_id = self.tokenizer.eos_token_id or 0
+
+        input_ids, attention_mask = [], []
+        for ids in encoded:
+            pad_len = max_seq_len - len(ids)
+            input_ids.append(ids + [pad_token_id] * pad_len)
+            attention_mask.append([1] * len(ids) + [0] * pad_len)
+
+        target_device = input_features.device
+        data = {
+            "input_ids": torch.tensor(input_ids, dtype=torch.long, device=target_device),
+            "attention_mask": torch.tensor(attention_mask, dtype=torch.long, device=target_device),
+            "input_features": input_features,
+            "audio_feature_lengths": audio_feature_lengths,
+            "audio_chunk_mapping": audio_chunk_mapping,
+        }
+        return BatchFeature(data=data, tensor_type=return_tensors)
+
+
+__all__ = ["MossTranscribeDiarizeProcessor"]

processor_config.json

@@ -0,0 +1,10 @@
+{
+  "processor_class": "MossTranscribeDiarizeProcessor",
+  "auto_map": {
+    "AutoProcessor": "processing_moss_transcribe_diarize.MossTranscribeDiarizeProcessor"
+  },
+  "audio_tokens_per_second": 12.5,
+  "audio_merge_size": 4,
+  "time_marker_every_seconds": 5,
+  "enable_time_marker": true
+}

special_tokens_map.json

@@ -0,0 +1,130 @@
+{
+  "additional_special_tokens": [
+    {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|audio_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|audio_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|audio_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    }
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd1921fc462f344d58efbf99d5af3806cffb8697edb584906efe27e26ac36fdf
+size 11423348

tokenizer_config.json

@@ -0,0 +1,267 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151665": {
+      "content": "<tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151666": {
+      "content": "</tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151667": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151668": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151669": {
+      "content": "<|audio_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151670": {
+      "content": "<|audio_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151671": {
+      "content": "<|audio_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>",
+    "<|audio_start|>",
+    "<|audio_end|>",
+    "<|audio_pad|>"
+  ],
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "fix_mistral_regex": true,
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}