[model] support MiMo-V2-Flash model (#9637)

README.md

@@ -313,6 +313,7 @@ Read technical notes:
 | [LLaVA-NeXT](https://huggingface.co/llava-hf)                     | 7B/8B/13B/34B/72B/110B           | llava_next           |
 | [LLaVA-NeXT-Video](https://huggingface.co/llava-hf)               | 7B/34B                           | llava_next_video     |
 | [MiMo](https://huggingface.co/XiaomiMiMo)                         | 7B                               | mimo                 |
+| [MiMo-v2](https://huggingface.co/XiaomiMiMo)                      | 309B                             | mimo_v2              |
 | [MiniCPM 1-4.1](https://huggingface.co/openbmb)                   | 0.5B/1B/2B/4B/8B                 | cpm/cpm3/cpm4        |
 | [MiniCPM-o-2.6/MiniCPM-V-2.6](https://huggingface.co/openbmb)     | 8B                               | minicpm_o/minicpm_v  |
 | [Ministral(3)/Mistral-Nemo](https://huggingface.co/mistralai)     | 3B/8B/12B/14B                    | ministral/ministral3 |

README_zh.md

@@ -315,6 +315,7 @@ https://github.com/user-attachments/assets/43b700c6-a178-41db-b1f8-8190a5d3fcfc
 | [LLaVA-NeXT](https://huggingface.co/llava-hf)                     | 7B/8B/13B/34B/72B/110B           | llava_next           |
 | [LLaVA-NeXT-Video](https://huggingface.co/llava-hf)               | 7B/34B                           | llava_next_video     |
 | [MiMo](https://huggingface.co/XiaomiMiMo)                         | 7B                               | mimo                 |
+| [MiMo-v2](https://huggingface.co/XiaomiMiMo)                      | 309B                             | mimo_v2              |
 | [MiniCPM 1-4.1](https://huggingface.co/openbmb)                   | 0.5B/1B/2B/4B/8B                 | cpm/cpm3/cpm4        |
 | [MiniCPM-o-2.6/MiniCPM-V-2.6](https://huggingface.co/openbmb)     | 8B                               | minicpm_o/minicpm_v  |
 | [Ministral(3)/Mistral-Nemo](https://huggingface.co/mistralai)     | 3B/8B/12B/14B                    | ministral/ministral3 |

src/llamafactory/data/template.py

@@ -1610,6 +1610,26 @@ register_template(
     template_class=ReasoningTemplate,
 )
 
+
+# copied from qwen template
+register_template(
+    name="mimo_v2",
+    format_user=StringFormatter(slots=["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]),
+    format_assistant=StringFormatter(slots=["{{content}}<|im_end|>\n"]),
+    format_system=StringFormatter(slots=["<|im_start|>system\n{{content}}<|im_end|>\n"]),
+    format_function=FunctionFormatter(slots=["{{content}}<|im_end|>\n"], tool_format="qwen"),
+    format_observation=StringFormatter(
+        slots=["<|im_start|>user\n<tool_response>\n{{content}}\n</tool_response><|im_end|>\n<|im_start|>assistant\n"]
+    ),
+    format_tools=ToolFormatter(tool_format="qwen"),
+    default_system="You are MiMo, a helpful AI assistant engineered by Xiaomi.",
+    stop_words=["<|im_end|>"],
+    replace_eos=True,
+    thought_words=("<think>", "</think>"),
+    template_class=ReasoningTemplate,
+)
+
+
 # copied from qwen2vl
 register_template(
     name="mimo_vl",

src/llamafactory/extras/constants.py

@@ -1803,6 +1803,21 @@ register_model_group(
 )
 
 
+register_model_group(
+    models={
+        "MiMo-V2-Flash-Base": {
+            DownloadSource.DEFAULT: "XiaomiMiMo/MiMo-V2-Flash-Base",
+            DownloadSource.MODELSCOPE: "XiaomiMiMo/MiMo-V2-Flash-Base",
+        },
+        "MiMo-V2-Flash": {
+            DownloadSource.DEFAULT: "XiaomiMiMo/MiMo-V2-Flash",
+            DownloadSource.MODELSCOPE: "XiaomiMiMo/MiMo-V2-Flash",
+        },
+    },
+    template="mimo_v2",
+)
+
+
 register_model_group(
     models={
         "MiMo-7B-VL-RL": {
@@ -1827,7 +1842,7 @@ register_model_group(
         },
         "MiMo-VL-7B-SFT-2508": {
             DownloadSource.DEFAULT: "XiaomiMiMo/MiMo-VL-7B-SFT-2508",
-            DownloadSource.DEFAULT: "XiaomiMiMo/MiMo-VL-7B-SFT-2508",
+            DownloadSource.MODELSCOPE: "XiaomiMiMo/MiMo-VL-7B-SFT-2508",
         },
     },
     template="qwen2_vl",

src/llamafactory/model/model_utils/quantization.py

@@ -113,6 +113,7 @@ def configure_quantization(
         if quant_method == QuantizationMethod.FP8:
             quant_config = FineGrainedFP8Config(dequantize=True)
             init_kwargs["quantization_config"] = quant_config
+            init_kwargs["ignore_mismatched_sizes"] = True
 
         quant_bits = quantization_config.get("bits", "?")
         logger.info_rank0(f"Loading {quant_bits}-bit {quant_method.upper()}-quantized model.")