Add files using upload-large-folder tool

audio_tokenizer/chat_template.jinja

@@ -0,0 +1,120 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0].role == 'system' %}
+        {%- if messages[0].content is string %}
+            {{- messages[0].content }}
+        {%- else %}
+            {%- for content in messages[0].content %}
+                {%- if content.type == 'audio' %}
+                    {{- ("<|sosp|>" + (content.meta | tojson) + "<|eosp|>") }}
+                {%- elif content.type == 'text' %}
+                    {{- content.text }}
+                {%- endif %}
+            {%- endfor %}
+        {%- endif %}
+    {%- endif %}
+    {{- '\n\n' }}
+    {{- "# 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' }}
+        {%- if messages[0].content is string %}
+            {{- messages[0].content }}
+        {%- else %}
+            {%- for content in messages[0].content %}
+                {%- if content.type == 'audio' %}
+                    {{- ("<|sosp|>" + (content.meta | tojson) + "<|eosp|>") }}
+                {%- elif content.type == 'text' %}
+                    {{- content.text }}
+                {%- endif %}
+            {%- endfor %}
+        {%- endif %}
+        {{- '\n<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1, assistant_is_last=false) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
+        {%- set ns.multi_step_tool = false %}
+        {%- set ns.last_query_index = index %}
+    {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+    {%- if message.content is string %}
+        {%- set content = message.content %}
+    {%- else %}
+        {%- set content = namespace(text="") %}
+        {%- for mcontent in message.content %}
+            {%- if mcontent.type == 'audio' %}
+                {%- set content.text = content.text~("<|sosp|>" + (mcontent.meta | tojson) + "<|eosp|>") %}
+            {%- elif mcontent.type == 'text' %}
+                {%- set content.text = content.text~mcontent.text %}
+            {%- endif %}
+        {%- endfor %}
+        {%- set content = content.text %}
+    {%- endif %}
+    {%- 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 %}
+        {%- if loop.last %}
+            {%- set ns.assistant_is_last = true %}
+        {%- else %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- 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 and not ns.assistant_is_last %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if audio_output  %}
+        {{- '<|sostm|>'}}
+    {%- elif not enable_thinking %}
+        {{- '<think>\n\n</think>\n' }}
+    {%- endif %}
+{%- endif %}

audio_tokenizer/config.json

@@ -0,0 +1,68 @@
+{
+  "max_audio_seconds": 300,
+  "stride_size": 2,
+  "avg_pooler": 2,
+  "d_model": 1024,
+  "scale_embedding": false,
+  "kernel_size": 3,
+  "activation_function": "gelu",
+  "encoder_layers": 24,
+  "encoder_skip_layer_id": 3,
+  "encoder_attention_heads": 16,
+  "encoder_ffn_dim": 4096,
+  "encoder_causal": true,
+  "encoder_attn_window_size": [
+    128,
+    0
+  ],
+  "decoder_layers": 24,
+  "decoder_attention_heads": 16,
+  "decoder_ffn_dim": 4096,
+  "decoder_kernel_size": 3,
+  "decoder_stride_size": 2,
+  "decoder_causal": true,
+  "decoder_attn_window_size": [
+    128,
+    0
+  ],
+  "nfft": 960,
+  "n_mels": 128,
+  "sampling_rate": 24000,
+  "hop_length": 240,
+  "window_size": 960,
+  "vocoder_padding": "same",
+  "fmin": 0,
+  "fmax": null,
+  "num_quantizers": 20,
+  "codebook_size": [
+    1024,
+    1024,
+    256,
+    128,
+    128,
+    128,
+    128,
+    128,
+    128,
+    128,
+    128,
+    128,
+    128,
+    128,
+    128,
+    128,
+    128,
+    128,
+    128,
+    128
+  ],
+  "threshold_ema_dead_code": 2,
+  "position_embedding_type": "rope",
+  "rope_theta": 10000,
+  "rope_type": "default",
+  "ln_type": "LayerNorm",
+  "use_istft_only": true,
+  "hybrid_attention": true,
+  "hybrid_block_size": 8,
+  "swa_per_block": 2
+}

audio_tokenizer/generation_config.json

@@ -0,0 +1,9 @@
+{
+  "do_sample": true,
+  "temperature": 0.6,
+  "top_k": -1,
+  "top_p": 0.95,
+  "audio_temperature": 0.9,
+  "audio_top_k": -1,
+  "audio_top_p": 0.95
+}

audio_tokenizer/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": "<|mimo_audio_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151666": {
+      "content": "<|mimo_audio_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "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_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151670": {
+      "content": "<|mimo_video_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151671": {
+      "content": "<|mimo_video_end|>",
+      "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_pad|>",
+    "<|mimo_audio_start|>",
+    "<|mimo_audio_end|>",
+    "<|mimo_video_start|>",
+    "<|mimo_video_end|>"
+  ],
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

config.json

@@ -0,0 +1,369 @@
+{
+  "architectures": [
+    "MiMoV2ForCausalLM"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_mimo_v2.MiMoV2Config"
+  },
+  "attention_bias": false,
+  "attention_chunk_size": 128,
+  "attention_dropout": 0.0,
+  "attention_projection_layout": "fused_qkv",
+  "attention_value_scale": 0.707,
+  "add_full_attention_sink_bias": false,
+  "add_swa_attention_sink_bias": true,
+  "audio_config": {
+    "add_post_norm": true,
+    "audio_channels": 20,
+    "audio_segment_size": 6000,
+    "group_size": 4,
+    "input_full_attention": true,
+    "input_local_attn_heads": 16,
+    "input_local_dim": 1024,
+    "input_local_head_dim": 64,
+    "input_local_hidden_dropout": 0.0,
+    "input_local_intermediate_size": 4096,
+    "input_local_layers": 6,
+    "out_hidden_size": 4096,
+    "partial_rotary_factor": 1.0,
+    "projection_layers": 2,
+    "rope_theta": 640000,
+    "speech_vocab_size": "1280",
+    "speech_zeroemb_idx": "1024",
+    "tokenizer_version": "v2"
+  },
+  "dtype": "bfloat16",
+  "eos_token_id": 151645,
+  "head_dim": 192,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "hybrid_block_size": null,
+  "hybrid_layer_pattern": [
+    0,
+    1,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    1,
+    1,
+    0,
+    1,
+    1,
+    1,
+    1,
+    1,
+    0
+  ],
+  "image_token_id": 151655,
+  "initializer_range": 0.02,
+  "intermediate_size": 16384,
+  "layernorm_epsilon": 1e-05,
+  "max_position_embeddings": 1048576,
+  "model_type": "mimo_v2",
+  "moe_intermediate_size": 2048,
+  "moe_layer_freq": [
+    0,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1
+  ],
+  "n_group": 1,
+  "n_routed_experts": 256,
+  "n_shared_experts": null,
+  "norm_topk_prob": true,
+  "num_attention_heads": 64,
+  "num_experts_per_tok": 8,
+  "num_hidden_layers": 48,
+  "num_key_value_heads": 4,
+  "pad_token_id": 151643,
+  "partial_rotary_factor": 0.334,
+  "processor_config": {
+    "audio_avg_pooler": 2,
+    "audio_channels": 20,
+    "audio_end_token_id": 151674,
+    "audio_fmax": null,
+    "audio_fmin": 0,
+    "audio_group_size": 4,
+    "audio_hop_length": 240,
+    "audio_input_id_per_second": 25.0,
+    "audio_kernel_size": 3,
+    "audio_n_mels": 128,
+    "audio_nfft": 960,
+    "audio_sampling_rate": 24000,
+    "audio_segment_size": 6000,
+    "audio_start_token_id": 151673,
+    "audio_stride_size": 2,
+    "audio_token_id": 151669,
+    "audio_window_size": 960,
+    "audio_zeroemb_idx": [
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024,
+      1024
+    ],
+    "fps": 1.0,
+    "image_max_pixels": 8388608,
+    "image_min_pixels": 8192,
+    "image_token_id": 151655,
+    "max_frames": 3600,
+    "merge_size": 2,
+    "min_frames": null,
+    "num_frames": null,
+    "pad_token_id": 151643,
+    "patch_size": 16,
+    "rope_type": "rope",
+    "temporal_compression_ratio": 1,
+    "temporal_patch_size": 2,
+    "use_per_grid_t_timestamps": false,
+    "use_video_timestamps": true,
+    "video_audio_interleave_length": 0.0,
+    "video_end_token_id": 151671,
+    "video_max_pixels": 8388608,
+    "video_min_pixels": 8192,
+    "video_process_num_threads": 16,
+    "video_start_token_id": 151670,
+    "video_token_id": 151656,
+    "video_tokens_per_second": 2,
+    "video_total_max_pixels": 268435456,
+    "vision_end_token_id": 151653,
+    "vision_start_token_id": 151652
+  },
+  "quantization_config": {
+    "activation_scheme": "dynamic",
+    "fmt": "e4m3",
+    "ignored_layers": [
+      "model.layers.0.self_attn.o_proj",
+      "model.layers.1.self_attn.o_proj",
+      "model.layers.2.self_attn.o_proj",
+      "model.layers.3.self_attn.o_proj",
+      "model.layers.4.self_attn.o_proj",
+      "model.layers.5.self_attn.o_proj",
+      "model.layers.6.self_attn.o_proj",
+      "model.layers.7.self_attn.o_proj",
+      "model.layers.8.self_attn.o_proj",
+      "model.layers.9.self_attn.o_proj",
+      "model.layers.10.self_attn.o_proj",
+      "model.layers.11.self_attn.o_proj",
+      "model.layers.12.self_attn.o_proj",
+      "model.layers.13.self_attn.o_proj",
+      "model.layers.14.self_attn.o_proj",
+      "model.layers.15.self_attn.o_proj",
+      "model.layers.16.self_attn.o_proj",
+      "model.layers.17.self_attn.o_proj",
+      "model.layers.18.self_attn.o_proj",
+      "model.layers.19.self_attn.o_proj",
+      "model.layers.20.self_attn.o_proj",
+      "model.layers.21.self_attn.o_proj",
+      "model.layers.22.self_attn.o_proj",
+      "model.layers.23.self_attn.o_proj",
+      "model.layers.24.self_attn.o_proj",
+      "model.layers.25.self_attn.o_proj",
+      "model.layers.26.self_attn.o_proj",
+      "model.layers.27.self_attn.o_proj",
+      "model.layers.28.self_attn.o_proj",
+      "model.layers.29.self_attn.o_proj",
+      "model.layers.30.self_attn.o_proj",
+      "model.layers.31.self_attn.o_proj",
+      "model.layers.32.self_attn.o_proj",
+      "model.layers.33.self_attn.o_proj",
+      "model.layers.34.self_attn.o_proj",
+      "model.layers.35.self_attn.o_proj",
+      "model.layers.36.self_attn.o_proj",
+      "model.layers.37.self_attn.o_proj",
+      "model.layers.38.self_attn.o_proj",
+      "model.layers.39.self_attn.o_proj",
+      "model.layers.40.self_attn.o_proj",
+      "model.layers.41.self_attn.o_proj",
+      "model.layers.42.self_attn.o_proj",
+      "model.layers.43.self_attn.o_proj",
+      "model.layers.44.self_attn.o_proj",
+      "model.layers.45.self_attn.o_proj",
+      "model.layers.46.self_attn.o_proj",
+      "model.layers.47.self_attn.o_proj",
+      "model.decoder.self_attn.o_proj"
+    ],
+    "quant_method": "fp8",
+    "store_dtype": "fp8",
+    "weight_block_size": [
+      128,
+      128
+    ]
+  },
+  "rope_scaling": {
+    "type": "default",
+    "rope_type": "default"
+  },
+  "rope_theta": 10000000,
+  "routed_scaling_factor": null,
+  "scoring_func": "sigmoid",
+  "sliding_window": 128,
+  "sliding_window_size": 128,
+  "swa_head_dim": 192,
+  "swa_num_attention_heads": 64,
+  "swa_num_key_value_heads": 8,
+  "swa_rope_theta": 10000,
+  "swa_v_head_dim": 128,
+  "tie_word_embeddings": false,
+  "topk_group": 1,
+  "topk_method": "noaux_tc",
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.40.1",
+  "use_cache": true,
+  "v_head_dim": 128,
+  "video_token_id": 151656,
+  "vision_config": {
+    "depth": 28,
+    "fullatt_block_indexes": [
+      0,
+      9,
+      18,
+      27
+    ],
+    "hidden_act": "silu",
+    "hidden_size": 1280,
+    "in_chans": 3,
+    "intermediate_size": 4608,
+    "num_heads": 32,
+    "num_key_value_heads": 8,
+    "num_query_groups": 4,
+    "out_hidden_size": 4096,
+    "patch_merger_conv": false,
+    "patch_merger_linear": false,
+    "patch_size": 16,
+    "spatial_merge_size": 2,
+    "spatial_patch_size": 16,
+    "temporal_patch_size": 2,
+    "tokens_per_second": 2,
+    "use_sink": true,
+    "visual_token_window_size": 64,
+    "vit_window_attn_types": [
+      -1,
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      -1,
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      -1,
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      -1
+    ],
+    "window_size": 128
+  },
+  "vision_end_token_id": 151653,
+  "vision_model_type": "mimovl",
+  "vision_start_token_id": 151652,
+  "vocab_size": 152576
+}

configuration_mimo_v2.py

@@ -0,0 +1,212 @@
+# coding=utf-8
+#
+# Copyright 2026 Xiaomi Corporation.
+# Copyright 2026 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.modeling_rope_utils import rope_config_validation
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+
+_MIMOV2_ATTENTION_PROJECTION_LAYOUTS = {"split", "fused_qkv"}
+
+_MIMOV2_SPLIT_TP_PLAN = {
+    "layers.*.self_attn.q_proj": "colwise",
+    "layers.*.self_attn.k_proj": "colwise",
+    "layers.*.self_attn.v_proj": "colwise",
+    "layers.*.self_attn.o_proj": "rowwise",
+    "layers.*.mlp.gate_proj": "colwise",
+    "layers.*.mlp.up_proj": "colwise",
+    "layers.*.mlp.down_proj": "rowwise",
+}
+
+_MIMOV2_FUSED_QKV_TP_PLAN = {
+    "layers.*.self_attn.qkv_proj": "colwise",
+    "layers.*.self_attn.o_proj": "rowwise",
+    "layers.*.mlp.gate_proj": "colwise",
+    "layers.*.mlp.up_proj": "colwise",
+    "layers.*.mlp.down_proj": "rowwise",
+}
+
+_MIMOV2_PP_PLAN = {
+    "embed_tokens": (["input_ids"], ["inputs_embeds"]),
+    "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
+    "norm": (["hidden_states"], ["hidden_states"]),
+}
+
+
+class MiMoV2Config(PretrainedConfig):
+
+    model_type = "mimo_v2"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    base_model_tp_plan = _MIMOV2_SPLIT_TP_PLAN
+    base_model_pp_plan = _MIMOV2_PP_PLAN
+
+    attribute_map = {
+        "num_local_experts": "n_routed_experts",
+    }
+
+    def __init__(
+        self,
+        vocab_size=151936,
+        hidden_size=4096,
+        intermediate_size=22016,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=32,
+        hidden_act="silu",
+        max_position_embeddings=32768,
+        initializer_range=0.02,
+        layernorm_epsilon=1e-6,
+        use_cache=True,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        attention_dropout=0.0,
+        attention_bias=False,
+        attention_value_scale=None,
+        head_dim=None,
+        v_head_dim=None,
+        swa_num_attention_heads=None,
+        swa_num_key_value_heads=None,
+        swa_head_dim=None,
+        swa_v_head_dim=None,
+        swa_rope_theta=None,
+        sliding_window=None,
+        sliding_window_size=None,
+        add_full_attention_sink_bias=False,
+        add_swa_attention_sink_bias=False,
+        hybrid_block_size=None,
+        hybrid_layer_pattern=None,
+        partial_rotary_factor=1.0,
+        n_routed_experts=None,
+        moe_intermediate_size=None,
+        num_experts_per_tok=None,
+        routed_scaling_factor=None,
+        scoring_func="sigmoid",
+        topk_method="noaux_tc",
+        n_group=None,
+        topk_group=None,
+        norm_topk_prob=True,
+        moe_layer_freq=None,
+        attention_projection_layout=None,
+        **kwargs,
+    ):
+        rope_parameters = kwargs.pop("rope_parameters", None)
+        if rope_scaling is None and rope_parameters is not None:
+            rope_scaling = rope_parameters
+
+        # NOTE: DO NOT auto-fill attention_projection_layout to "split" when unset.
+        # sglang's server_args (MIMO_V2_MODEL_ARCHS branch) rejects any
+        # non-"fused_qkv" value that's explicitly present on the config; leaving
+        # the attribute absent triggers the None early-return path.
+        if attention_projection_layout is not None and attention_projection_layout not in _MIMOV2_ATTENTION_PROJECTION_LAYOUTS:
+            raise ValueError(f"Unsupported MiMoV2 attention projection layout: {attention_projection_layout}")
+
+        if attention_projection_layout is not None:
+            self.attention_projection_layout = attention_projection_layout
+        self.base_model_tp_plan = (
+            _MIMOV2_FUSED_QKV_TP_PLAN.copy()
+            if attention_projection_layout == "fused_qkv"
+            else _MIMOV2_SPLIT_TP_PLAN.copy()
+        )
+        self.base_model_pp_plan = _MIMOV2_PP_PLAN.copy()
+
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+        if num_attention_heads % num_key_value_heads != 0:
+            raise ValueError("num_attention_heads must be divisible by num_key_value_heads")
+
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.layernorm_epsilon = layernorm_epsilon
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+        self.attention_dropout = attention_dropout
+        self.attention_bias = attention_bias
+        self.attention_value_scale = attention_value_scale
+
+        self.head_dim = head_dim if head_dim is not None else hidden_size // num_attention_heads
+        self.v_head_dim = v_head_dim if v_head_dim is not None else self.head_dim
+        self.swa_num_attention_heads = (
+            swa_num_attention_heads if swa_num_attention_heads is not None else num_attention_heads
+        )
+        self.swa_num_key_value_heads = (
+            swa_num_key_value_heads if swa_num_key_value_heads is not None else num_key_value_heads
+        )
+        if self.swa_num_attention_heads % self.swa_num_key_value_heads != 0:
+            raise ValueError("swa_num_attention_heads must be divisible by swa_num_key_value_heads")
+        self.swa_head_dim = swa_head_dim if swa_head_dim is not None else self.head_dim
+        self.swa_v_head_dim = swa_v_head_dim if swa_v_head_dim is not None else self.swa_head_dim
+        self.swa_rope_theta = swa_rope_theta if swa_rope_theta is not None else rope_theta
+
+        if sliding_window is None:
+            sliding_window = sliding_window_size
+        self.sliding_window = sliding_window
+        self.sliding_window_size = sliding_window_size if sliding_window_size is not None else sliding_window
+        self.add_full_attention_sink_bias = add_full_attention_sink_bias
+        self.add_swa_attention_sink_bias = add_swa_attention_sink_bias
+
+        if hybrid_block_size is not None and hybrid_layer_pattern is None:
+            hybrid_layer_pattern = [0 if ((i + 1) % hybrid_block_size == 0) else 1 for i in range(num_hidden_layers)]
+        elif hybrid_layer_pattern is None:
+            hybrid_layer_pattern = [0] * num_hidden_layers
+        if len(hybrid_layer_pattern) != num_hidden_layers:
+            raise ValueError("hybrid_layer_pattern length must match num_hidden_layers")
+        self.hybrid_block_size = hybrid_block_size
+        self.hybrid_layer_pattern = hybrid_layer_pattern
+
+        self.partial_rotary_factor = partial_rotary_factor
+
+        self.n_routed_experts = n_routed_experts
+        self.moe_intermediate_size = moe_intermediate_size if moe_intermediate_size is not None else intermediate_size
+        self.num_experts_per_tok = num_experts_per_tok
+        self.routed_scaling_factor = routed_scaling_factor
+        self.scoring_func = scoring_func
+        self.topk_method = topk_method
+        self.n_group = n_group
+        self.topk_group = topk_group
+        self.norm_topk_prob = norm_topk_prob
+        if isinstance(moe_layer_freq, int):
+            moe_layer_freq = [moe_layer_freq > 0 and i % moe_layer_freq == 0 for i in range(num_hidden_layers)]
+        elif moe_layer_freq is None:
+            moe_layer_freq = [False] * num_hidden_layers
+        if len(moe_layer_freq) != num_hidden_layers:
+            raise ValueError("moe_layer_freq length must match num_hidden_layers")
+        self.moe_layer_freq = moe_layer_freq
+
+        if self.rope_scaling is not None and "type" in self.rope_scaling:
+            self.rope_scaling["rope_type"] = self.rope_scaling["type"]
+        rope_config_validation(self)
+
+        super().__init__(
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+__all__ = ["MiMoV2Config"]

dflash/config.json

@@ -0,0 +1,48 @@
+{
+    "architectures": [
+        "DFlashDraftModel"
+    ],
+    "model_type": "qwen3",
+    "auto_map": {
+        "AutoModel": "dflash.DFlashDraftModel"
+    },
+    "hidden_size": 4096,
+    "intermediate_size": 16384,
+    "num_hidden_layers": 5,
+    "num_attention_heads": 64,
+    "num_key_value_heads": 8,
+    "head_dim": 128,
+    "v_head_dim": 128,
+    "partial_rotary_factor": 0.5,
+    "block_size": 8,
+    "dflash_config": {
+        "target_layer_ids": [
+            0,
+            11,
+            23,
+            35,
+            47
+        ],
+        "mask_token_id": 151675,
+        "num_anchors": 4096,
+        "block_size": 8,
+        "loss_decay_gamma": 7.0,
+        "use_swa": true,
+        "swa_window_size": 1024,
+        "backbone_rotary_base": 5000000,
+        "attention_value_scale": 0.612,
+        "attention_sink_bias": true
+    },
+    "num_target_layers": 48,
+    "vocab_size": 152576,
+    "max_position_embeddings": 262144,
+    "rope_theta": 10000,
+    "sliding_window": 1024,
+    "rms_norm_eps": 1e-06,
+    "torch_dtype": "bfloat16",
+    "hidden_act": "silu",
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "tie_word_embeddings": false,
+    "use_cache": true
+}

dflash/dflash.py

@@ -0,0 +1,379 @@
+from typing import Callable, Optional
+
+import torch
+from torch import nn
+from transformers import DynamicCache
+from transformers.cache_utils import Cache
+from transformers.modeling_outputs import CausalLMOutputWithPast
+from transformers.models.qwen3.modeling_qwen3 import (
+    ALL_ATTENTION_FUNCTIONS,
+    FlashAttentionKwargs,
+    GradientCheckpointingLayer,
+    Qwen3Config,
+    Qwen3MLP,
+    Qwen3PreTrainedModel,
+    Qwen3RMSNorm,
+    Qwen3RotaryEmbedding,
+    eager_attention_forward,
+    rotate_half,
+)
+from typing_extensions import Tuple, Unpack
+
+
+def sample(logits: torch.Tensor, temperature: float = 0.0) -> torch.Tensor:
+    if temperature < 1e-5:
+        return torch.argmax(logits, dim=-1)
+    bsz, seq_len, vocab_size = logits.shape
+    logits = logits.view(-1, vocab_size)
+    logits = logits / temperature
+    probs = torch.softmax(logits, dim=-1)
+    return torch.multinomial(probs, num_samples=1).view(bsz, seq_len)
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+    q_len = q.size(-2)
+    q_embed = (q * cos[..., -q_len:, :]) + (rotate_half(q) * sin[..., -q_len:, :])
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+class Qwen3DFlashAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: Qwen3Config, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.head_dim = getattr(
+            config, "head_dim", config.hidden_size // config.num_attention_heads
+        )
+        self.num_key_value_groups = (
+            config.num_attention_heads // config.num_key_value_heads
+        )
+        self.scaling = self.head_dim**-0.5
+        self.attention_dropout = config.attention_dropout
+        self.is_causal = False
+        self.q_proj = nn.Linear(
+            config.hidden_size,
+            config.num_attention_heads * self.head_dim,
+            bias=config.attention_bias,
+        )
+        self.k_proj = nn.Linear(
+            config.hidden_size,
+            config.num_key_value_heads * self.head_dim,
+            bias=config.attention_bias,
+        )
+        self.v_proj = nn.Linear(
+            config.hidden_size,
+            config.num_key_value_heads * self.head_dim,
+            bias=config.attention_bias,
+        )
+        self.o_proj = nn.Linear(
+            config.num_attention_heads * self.head_dim,
+            config.hidden_size,
+            bias=config.attention_bias,
+        )
+        self.q_norm = Qwen3RMSNorm(self.head_dim, eps=config.rms_norm_eps)
+        self.k_norm = Qwen3RMSNorm(self.head_dim, eps=config.rms_norm_eps)
+        self.sliding_window = (
+            config.sliding_window
+            if config.layer_types[layer_idx] == "sliding_attention"
+            else None
+        )
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        target_hidden: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor],
+        attention_mask: Optional[torch.Tensor],
+        past_key_values: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
+        bsz, q_len = hidden_states.shape[:-1]
+        ctx_len = target_hidden.shape[1]
+        q = self.q_proj(hidden_states)
+        q = q.view(bsz, q_len, -1, self.head_dim)
+        q = self.q_norm(q).transpose(1, 2)
+        k_ctx = self.k_proj(target_hidden)
+        k_noise = self.k_proj(hidden_states)
+        v_ctx = self.v_proj(target_hidden)
+        v_noise = self.v_proj(hidden_states)
+        k = torch.cat([k_ctx, k_noise], dim=1).view(
+            bsz, ctx_len + q_len, -1, self.head_dim
+        )
+        v = torch.cat([v_ctx, v_noise], dim=1).view(
+            bsz, ctx_len + q_len, -1, self.head_dim
+        )
+        k = self.k_norm(k).transpose(1, 2)
+        v = v.transpose(1, 2)
+        cos, sin = position_embeddings
+        q, k = apply_rotary_pos_emb(q, k, cos, sin)
+        if past_key_values is not None:
+            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
+            k, v = past_key_values.update(k, v, self.layer_idx, cache_kwargs)
+        attn_fn: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            attn_fn = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+        attn_output, attn_weights = attn_fn(
+            self,
+            q,
+            k,
+            v,
+            attention_mask,
+            dropout=0.0 if not self.training else self.attention_dropout,
+            scaling=self.scaling,
+            sliding_window=self.sliding_window,
+            **kwargs,
+        )
+        attn_output = attn_output.reshape(bsz, q_len, -1)
+        attn_output = self.o_proj(attn_output)
+        return attn_output, attn_weights
+
+
+class Qwen3DFlashDecoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config: Qwen3Config, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.self_attn = Qwen3DFlashAttention(config=config, layer_idx=layer_idx)
+        self.mlp = Qwen3MLP(config)
+        self.input_layernorm = Qwen3RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = Qwen3RMSNorm(
+            config.hidden_size, eps=config.rms_norm_eps
+        )
+
+    def forward(
+        self,
+        target_hidden: Optional[torch.Tensor] = None,
+        hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+        cache_position: Optional[torch.LongTensor] = None,
+        position_embeddings: Optional[
+            Tuple[torch.Tensor, torch.Tensor]
+        ] = None,  # necessary, but kept here for BC
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> Tuple[
+        torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]
+    ]:
+        residual = hidden_states
+        hidden_states = self.input_layernorm(hidden_states)
+        hidden_states = self.self_attn(
+            hidden_states=hidden_states,
+            target_hidden=target_hidden,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            position_embeddings=position_embeddings,
+            **kwargs,
+        )[0]
+        hidden_states = residual + hidden_states
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+        return hidden_states
+
+
+def build_target_layer_ids(num_target_layers: int, num_draft_layers: int):
+    if num_draft_layers == 1:
+        return [(num_target_layers // 2)]
+    start = 1
+    end = num_target_layers - 3
+    span = end - start
+    target_layer_ids = [
+        int(round(start + (i * span) / (num_draft_layers - 1)))
+        for i in range(num_draft_layers)
+    ]
+    return target_layer_ids
+
+
+def extract_context_feature(
+    hidden_states: list[torch.Tensor],
+    layer_ids: Optional[list[int]],
+) -> torch.Tensor:
+    offset = 1
+    selected_states = []
+    for layer_id in layer_ids:
+        selected_states.append(hidden_states[layer_id + offset])
+    target_hidden = torch.cat(selected_states, dim=-1)
+    return target_hidden
+
+
+class DFlashDraftModel(Qwen3PreTrainedModel):
+    config_class = Qwen3Config
+    _no_split_modules = ["Qwen3DFlashDecoderLayer"]
+
+    def __init__(self, config) -> None:
+        super().__init__(config)
+        self.config = config
+        self.layers = nn.ModuleList(
+            [
+                Qwen3DFlashDecoderLayer(config, layer_idx)
+                for layer_idx in range(config.num_hidden_layers)
+            ]
+        )
+        dflash_config = getattr(config, "dflash_config", {}) or {}
+        self.target_layer_ids = dflash_config.get(
+            "target_layer_ids",
+            build_target_layer_ids(config.num_target_layers, config.num_hidden_layers),
+        )
+        self.norm = Qwen3RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = Qwen3RotaryEmbedding(config)
+        self.fc = nn.Linear(
+            len(self.target_layer_ids) * config.hidden_size,
+            config.hidden_size,
+            bias=False,
+        )
+        self.hidden_norm = Qwen3RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.block_size = config.block_size
+        self.mask_token_id = dflash_config.get("mask_token_id", None)
+        self.post_init()
+
+    def forward(
+        self,
+        position_ids: torch.LongTensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        noise_embedding: Optional[torch.Tensor] = None,
+        target_hidden: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Cache] = None,
+        use_cache: bool = False,
+        **kwargs,
+    ) -> CausalLMOutputWithPast:
+        hidden_states = noise_embedding
+        target_hidden = self.hidden_norm(self.fc(target_hidden))
+        position_embeddings = self.rotary_emb(hidden_states, position_ids)
+        for layer in self.layers:
+            hidden_states = layer(
+                hidden_states=hidden_states,
+                target_hidden=target_hidden,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_values,
+                use_cache=use_cache,
+                position_embeddings=position_embeddings,
+                **kwargs,
+            )
+        return self.norm(hidden_states)
+
+    @torch.inference_mode()
+    def spec_generate(
+        self,
+        target: nn.Module,
+        input_ids: torch.LongTensor,
+        max_new_tokens: int,
+        stop_token_ids: list[int],
+        temperature: float,
+    ):
+        self.eval()
+        num_input_tokens = input_ids.shape[1]
+        max_length = num_input_tokens + max_new_tokens
+
+        block_size = self.block_size
+        output_ids = torch.full(
+            (1, max_length + block_size),
+            self.mask_token_id,
+            dtype=torch.long,
+            device=target.device,
+        )
+        position_ids = torch.arange(
+            output_ids.shape[1], device=target.device
+        ).unsqueeze(0)
+
+        past_key_values_target = DynamicCache()
+        past_key_values_draft = DynamicCache()
+
+        # Prefill stage
+        output = target(
+            input_ids,
+            position_ids=position_ids[:, :num_input_tokens],
+            past_key_values=past_key_values_target,
+            use_cache=True,
+            logits_to_keep=1,
+            output_hidden_states=True,
+        )
+
+        output_ids[:, :num_input_tokens] = input_ids
+        output_ids[:, num_input_tokens : num_input_tokens + 1] = sample(
+            output.logits, temperature
+        )
+        target_hidden = extract_context_feature(
+            output.hidden_states, self.target_layer_ids
+        )
+
+        # Decode stage
+        acceptance_lengths = []
+        start = input_ids.shape[1]
+        while start < max_length:
+            block_output_ids = output_ids[:, start : start + block_size].clone()
+            block_position_ids = position_ids[:, start : start + block_size]
+            noise_embedding = target.model.embed_tokens(block_output_ids)
+            draft_logits = target.lm_head(
+                self(
+                    target_hidden=target_hidden,
+                    noise_embedding=noise_embedding,
+                    position_ids=position_ids[
+                        :, past_key_values_draft.get_seq_length() : start + block_size
+                    ],
+                    past_key_values=past_key_values_draft,
+                    use_cache=True,
+                    is_causal=False,
+                )[:, -block_size + 1 :, :]
+            )
+            past_key_values_draft.crop(start)
+            block_output_ids[:, 1:] = sample(draft_logits)
+
+            output = target(
+                block_output_ids,
+                position_ids=block_position_ids,
+                past_key_values=past_key_values_target,
+                use_cache=True,
+                output_hidden_states=True,
+            )
+
+            posterior = sample(output.logits, temperature)
+            acceptance_length = (
+                (block_output_ids[:, 1:] == posterior[:, :-1])
+                .cumprod(dim=1)
+                .sum(dim=1)[0]
+                .item()
+            )
+            output_ids[:, start : start + acceptance_length + 1] = block_output_ids[
+                :, : acceptance_length + 1
+            ]
+            output_ids[:, start + acceptance_length + 1] = posterior[
+                :, acceptance_length
+            ]
+            start += acceptance_length + 1
+            past_key_values_target.crop(start)
+            target_hidden = extract_context_feature(
+                output.hidden_states, self.target_layer_ids
+            )[:, : acceptance_length + 1, :]
+            acceptance_lengths.append(acceptance_length + 1)
+            if stop_token_ids is not None and any(
+                stop_token_id in output_ids[:, num_input_tokens:]
+                for stop_token_id in stop_token_ids
+            ):
+                break
+        output_ids = output_ids[:, :max_length]
+        output_ids = output_ids[:, output_ids[0] != self.mask_token_id]
+        if stop_token_ids is not None:
+            stop_token_ids = torch.tensor(stop_token_ids, device=output_ids.device)
+            stop_token_indices = torch.isin(
+                output_ids[0][num_input_tokens:], stop_token_ids
+            ).nonzero(as_tuple=True)[0]
+            if stop_token_indices.numel() > 0:
+                output_ids = output_ids[
+                    :, : num_input_tokens + stop_token_indices[0] + 1
+                ]
+
+        return output_ids

dflash/done_mtp

@@ -0,0 +1 @@
+1

dflash/mask_embedding.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a982aea3bb45e78bbe59bd3d624cfbc3f1261958e0f466e64f14b9e22377b724
+size 9882

dflash/model.safetensors.index.json

@@ -0,0 +1,70 @@
+{
+  "metadata": {
+    "total_size": 2936114304
+  },
+  "weight_map": {
+    "fc.weight": "dflash_draft_model.safetensors",
+    "hidden_norm.weight": "dflash_draft_model.safetensors",
+    "norm.weight": "dflash_draft_model.safetensors",
+    "layers.0.input_layernorm.weight": "dflash_draft_model.safetensors",
+    "layers.0.post_attention_layernorm.weight": "dflash_draft_model.safetensors",
+    "layers.0.self_attn.q_proj.weight": "dflash_draft_model.safetensors",
+    "layers.0.self_attn.k_proj.weight": "dflash_draft_model.safetensors",
+    "layers.0.self_attn.v_proj.weight": "dflash_draft_model.safetensors",
+    "layers.0.self_attn.o_proj.weight": "dflash_draft_model.safetensors",
+    "layers.0.self_attn.q_norm.weight": "dflash_draft_model.safetensors",
+    "layers.0.self_attn.k_norm.weight": "dflash_draft_model.safetensors",
+    "layers.0.self_attn.attention_sink_bias": "dflash_draft_model.safetensors",
+    "layers.0.mlp.gate_proj.weight": "dflash_draft_model.safetensors",
+    "layers.0.mlp.up_proj.weight": "dflash_draft_model.safetensors",
+    "layers.0.mlp.down_proj.weight": "dflash_draft_model.safetensors",
+    "layers.1.input_layernorm.weight": "dflash_draft_model.safetensors",
+    "layers.1.post_attention_layernorm.weight": "dflash_draft_model.safetensors",
+    "layers.1.self_attn.q_proj.weight": "dflash_draft_model.safetensors",
+    "layers.1.self_attn.k_proj.weight": "dflash_draft_model.safetensors",
+    "layers.1.self_attn.v_proj.weight": "dflash_draft_model.safetensors",
+    "layers.1.self_attn.o_proj.weight": "dflash_draft_model.safetensors",
+    "layers.1.self_attn.q_norm.weight": "dflash_draft_model.safetensors",
+    "layers.1.self_attn.k_norm.weight": "dflash_draft_model.safetensors",
+    "layers.1.self_attn.attention_sink_bias": "dflash_draft_model.safetensors",
+    "layers.1.mlp.gate_proj.weight": "dflash_draft_model.safetensors",
+    "layers.1.mlp.up_proj.weight": "dflash_draft_model.safetensors",
+    "layers.1.mlp.down_proj.weight": "dflash_draft_model.safetensors",
+    "layers.2.input_layernorm.weight": "dflash_draft_model.safetensors",
+    "layers.2.post_attention_layernorm.weight": "dflash_draft_model.safetensors",
+    "layers.2.self_attn.q_proj.weight": "dflash_draft_model.safetensors",
+    "layers.2.self_attn.k_proj.weight": "dflash_draft_model.safetensors",
+    "layers.2.self_attn.v_proj.weight": "dflash_draft_model.safetensors",
+    "layers.2.self_attn.o_proj.weight": "dflash_draft_model.safetensors",
+    "layers.2.self_attn.q_norm.weight": "dflash_draft_model.safetensors",
+    "layers.2.self_attn.k_norm.weight": "dflash_draft_model.safetensors",
+    "layers.2.self_attn.attention_sink_bias": "dflash_draft_model.safetensors",
+    "layers.2.mlp.gate_proj.weight": "dflash_draft_model.safetensors",
+    "layers.2.mlp.up_proj.weight": "dflash_draft_model.safetensors",
+    "layers.2.mlp.down_proj.weight": "dflash_draft_model.safetensors",
+    "layers.3.input_layernorm.weight": "dflash_draft_model.safetensors",
+    "layers.3.post_attention_layernorm.weight": "dflash_draft_model.safetensors",
+    "layers.3.self_attn.q_proj.weight": "dflash_draft_model.safetensors",
+    "layers.3.self_attn.k_proj.weight": "dflash_draft_model.safetensors",
+    "layers.3.self_attn.v_proj.weight": "dflash_draft_model.safetensors",
+    "layers.3.self_attn.o_proj.weight": "dflash_draft_model.safetensors",
+    "layers.3.self_attn.q_norm.weight": "dflash_draft_model.safetensors",
+    "layers.3.self_attn.k_norm.weight": "dflash_draft_model.safetensors",
+    "layers.3.self_attn.attention_sink_bias": "dflash_draft_model.safetensors",
+    "layers.3.mlp.gate_proj.weight": "dflash_draft_model.safetensors",
+    "layers.3.mlp.up_proj.weight": "dflash_draft_model.safetensors",
+    "layers.3.mlp.down_proj.weight": "dflash_draft_model.safetensors",
+    "layers.4.input_layernorm.weight": "dflash_draft_model.safetensors",
+    "layers.4.post_attention_layernorm.weight": "dflash_draft_model.safetensors",
+    "layers.4.self_attn.q_proj.weight": "dflash_draft_model.safetensors",
+    "layers.4.self_attn.k_proj.weight": "dflash_draft_model.safetensors",
+    "layers.4.self_attn.v_proj.weight": "dflash_draft_model.safetensors",
+    "layers.4.self_attn.o_proj.weight": "dflash_draft_model.safetensors",
+    "layers.4.self_attn.q_norm.weight": "dflash_draft_model.safetensors",
+    "layers.4.self_attn.k_norm.weight": "dflash_draft_model.safetensors",
+    "layers.4.self_attn.attention_sink_bias": "dflash_draft_model.safetensors",
+    "layers.4.mlp.gate_proj.weight": "dflash_draft_model.safetensors",
+    "layers.4.mlp.up_proj.weight": "dflash_draft_model.safetensors",
+    "layers.4.mlp.down_proj.weight": "dflash_draft_model.safetensors"
+  }
+}

generation_config.json

@@ -0,0 +1,9 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": false,
+  "eos_token_id": [151643, 151645, 151672],
+  "temperature": 1.0,
+  "top_p": 0.95,
+  "max_new_tokens": 2048,
+  "transformers_version": "4.37.0"
+}

preprocessor_config.json

@@ -0,0 +1,19 @@
+{
+  "min_pixels": 3136,
+  "max_pixels": 12845056,
+  "patch_size": 16,
+  "temporal_patch_size": 2,
+  "merge_size": 2,
+  "image_mean": [
+    0.48145466,
+    0.4578275,
+    0.40821073
+  ],
+  "image_std": [
+    0.26862954,
+    0.26130258,
+    0.27577711
+  ],
+  "image_processor_type": "Qwen2VLImageProcessor",
+  "processor_class": "Qwen2_5_VLProcessor"
+}

tokenizer_config.json

@@ -0,0 +1,293 @@
+{
+  "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_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151670": {
+      "content": "<|mimo_video_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151671": {
+      "content": "<|mimo_video_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151672": {
+      "content": "<|mimo_audio_eod|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151673": {
+      "content": "<|mimo_audio_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151674": {
+      "content": "<|mimo_audio_end|>",
+      "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_pad|>",
+    "<|mimo_video_start|>",
+    "<|mimo_video_end|>",
+    "<|mimo_audio_eod|>",
+    "<|mimo_audio_start|>",
+    "<|mimo_audio_end|>"
+  ],
+  "bos_token": null,
+  "chat_template": "{%- if not add_generation_prompt is defined -%}\n    {%- set add_generation_prompt = false -%}\n{%- endif -%}\n{%- if not enable_thinking is defined -%}\n    {%- set enable_thinking = true -%}\n{%- endif -%}\n{%- if not keep_all_reasoning is defined -%}\n    {%- set keep_all_reasoning = true -%}\n{%- endif -%}\n{%- macro render_extra_keys(json_dict, handled_keys) -%}\n    {%- if json_dict is mapping %}\n        {%- for json_key in json_dict if json_key not in handled_keys %}\n            {%- if json_dict[json_key] is mapping or (json_dict[json_key] is sequence and json_dict[json_key] is not string) %}\n                {{- '\\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | tojson | safe) ~ '</' ~ json_key ~ '>' }}\n            {%- else %}\n                {{-'\\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | string) ~ '</' ~ json_key ~ '>' }}\n            {%- endif %}\n        {%- endfor %}\n    {%- endif %}\n{%- endmacro -%}\n{%- macro render_content(message_content) -%}\n    {%- if message_content is string -%}\n        {{- message_content -}}\n    {%- else -%}\n        {%- for content in message_content -%}\n            {%- if content['type'] == 'image' or 'image' in content or 'image_url' in content -%}\n                {{- '<|vision_start|><|image_pad|><|vision_end|>' -}}\n            {%- elif content['type'] == 'audio' or 'audio' in content or 'audio_url' in content -%}\n                {{- '<|mimo_audio_start|><|audio_pad|><|mimo_audio_end|>' -}}\n            {%- elif content['type'] == 'video' or 'video' in content or 'video_url' in content -%}\n                {{- '<|vision_start|><|video_pad|><|vision_end|>' -}}\n            {%- elif 'text' in content -%}\n                {{- content['text'] -}}\n            {%- endif -%}\n        {%- endfor -%}\n    {%- endif -%}\n{%- endmacro -%}\n{%- if messages[0][\"role\"] == \"system\" %}\n    {%- set system_message = messages[0][\"content\"] %}\n    {%- set loop_messages = messages[1:] %}\n{%- else %}\n    {%- set loop_messages = messages %}\n{%- endif %}\n{%- set ns = namespace(last_user_index=-1) %}\n{%- for m in loop_messages %}\n    {%- if m.role == 'user' %}\n        {%- set ns.last_user_index = loop.index0 -%}\n    {%- endif %}\n{%- endfor %}\n{%- if not tools is defined %}\n    {%- set tools = [] %}\n{%- endif %}\n{%- if system_message is defined %}\n    {{- \"<|im_start|>system\\n\" + render_content(system_message) }}\n{%- else %}\n    {{- \"<|im_start|>system\\nYou are MiMo, a helpful AI assistant engineered by Xiaomi.\" }}\n{%- endif %}\n{%- if tools is iterable and tools | length > 0 %}\n    {{- \"\\n\\n# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou have access to the following functions:\\n\\n\" }}\n    {{- \"<tools>\" }}\n    {%- for tool in tools %}\n        {%- if tool.function is defined %}\n            {%- set tool = tool.function %}\n        {%- endif %}\n        {{- \"\\n<function>\\n<name>\" ~ tool.name ~ \"</name>\" }}\n        {%- if tool.description is defined %}\n            {{- '\\n<description>' ~ (tool.description | trim) ~ '</description>' }}\n        {%- endif %}\n        {{- '\\n<parameters>' }}\n        {%- if tool.parameters is defined and tool.parameters is mapping and tool.parameters.properties is defined and tool.parameters.properties is mapping %}\n            {%- for param_name, param_fields in tool.parameters.properties|items %}\n                {{- '\\n<parameter>' }}\n                {{- '\\n<name>' ~ param_name ~ '</name>' }}\n                {%- if param_fields.type is defined %}\n                    {{- '\\n<type>' ~ (param_fields.type | string) ~ '</type>' }}\n                {%- endif %}\n                {%- if param_fields.description is defined %}\n                    {{- '\\n<description>' ~ (param_fields.description | trim) ~ '</description>' }}\n                {%- endif %}\n                {%- set handled_keys = ['name', 'type', 'description'] %}\n                {{- render_extra_keys(param_fields, handled_keys) }}\n                {{- '\\n</parameter>' }}\n            {%- endfor %}\n        {%- endif %}\n        {%- set handled_keys = ['type', 'properties'] %}\n        {{- render_extra_keys(tool.parameters, handled_keys) }}\n        {{- '\\n</parameters>' }}\n        {%- set handled_keys = ['type', 'name', 'description', 'parameters'] %}\n        {{- render_extra_keys(tool, handled_keys) }}\n        {{- '\\n</function>' }}\n    {%- endfor %}\n    {{- \"\\n</tools>\" }}\n    {{- '\\n\\nFor each function call, output the function name and arguments in the following format:\\n<tool_call>\\n<function=example_function_name>\\n<parameter=example_parameter_1>value_1</parameter>\\n<parameter=example_parameter_2>This is the value for the second parameter\\nthat can span\\nmultiple lines</parameter>\\n</function>\\n</tool_call>\\n\\n<IMPORTANT>\\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\\n- DO NOT use function calls inside <think></think> tags.\\n- The value enclosed between parameter tags is preserved exactly as-is, including newlines and spaces.\\n</IMPORTANT>' }}\n{%- endif %}\n{{- '<|im_end|>' }}\n{%- for message in loop_messages %}\n    {%- if message.content is string %}\n        {%- set content = message.content %}\n    {%- else %}\n        {%- set content = render_content(message.content) %}\n    {%- endif %}\n    {%- if message.role == \"assistant\" %}\n        {%- if message.reasoning_content is string %}\n            {%- set reasoning_content = message.reasoning_content %}\n        {%- else %}\n            {%- set reasoning_content = '' %}\n            {%- if '</think>' in content %}\n                {%- set reasoning_content = content.split('</think>')[0].split('<think>')[-1] %}\n                {%- set content = content.split('</think>')[-1] %}\n            {%- endif %}\n        {%- endif %}\n        {%- if (keep_all_reasoning or loop.index0 > ns.last_user_index) and reasoning_content -%}\n            {{- '<|im_start|>' + message.role + '\\n<think>' + reasoning_content + '</think>' + content }}\n        {%- else %}\n            {{- '<|im_start|>' + message.role + '\\n<think></think>' + content }}\n        {%- endif %}\n        {%- if message.tool_calls is defined and message.tool_calls is iterable and message.tool_calls | length > 0 %}\n            {%- for tool_call in message.tool_calls %}\n                {%- if tool_call.function is defined %}\n                    {%- set tool_call = tool_call.function %}\n                {%- endif %}\n                {{- '<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                {%- if tool_call.arguments is defined %}\n                    {%- for args_name, args_value in tool_call.arguments|items %}\n                        {{- '<parameter=' + args_name + '>' }}\n                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}\n                        {{- args_value }}\n                        {{- '</parameter>\\n' }}\n                    {%- endfor %}\n                {%- endif %}\n                {{- '</function>\\n</tool_call>' }}\n            {%- endfor %}\n        {%- endif %}\n        {{- '<|im_end|>' }}\n    {%- elif message.role == \"user\" %}\n        {{- '<|im_start|>' + message.role + '\\n' + render_content(message.content) + '<|im_end|>' }}\n    {%- elif message.role == \"system\" %}\n        {{- '<|im_start|>' + message.role + '\\n' + render_content(message.content) + '<|im_end|>' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if loop.previtem and loop.previtem.role != \"tool\" %}\n            {{- '<|im_start|>tool\\n' }}\n        {%- endif %}\n        {{- '<tool_response>\\n' }}\n        {{- render_content(message.content) }}\n        {{- '\\n</tool_response>\\n' }}\n        {%- if not loop.last and loop.nextitem.role != \"tool\" %}\n            {{- '<|im_end|>' }}\n        {%- elif loop.last %}\n            {{- '<|im_end|>' }}\n        {%- endif %}\n    {%- else %}\n        {{- '<|im_start|>' + message.role + '\\n' + render_content(message.content) + '<|im_end|>' }}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n    {%- if not enable_thinking -%}\n        {{- '<think></think>' -}}\n    {%- else -%}\n        {{- '' -}}\n    {%- endif -%}\n{%- endif %}\n",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}