3. 模型支持列表

该版本支持的模型列表如下：

3.1. 模型功能支持列表

表 3.1 模型功能支持列表

模型类型	模型名	量化	编译	示例包	备注
ASR	FireRedASR	✓	✓	X
ASR	GLM-ASR-Nano-2512	✓	✓	✓
ASR	Qwen3-ASR	✓	✓	✓
ASR	Whisper-large-v3-turbo-0.8B	✓	✓	✓
ASR	Whisper-medium	✓	✓	✓
Autonomous Driving	YOLOP	✓	✓	✓
Backbone	EfficientNet	✓	✓	✓
Backbone	MobileNetV2	✓	✓	✓
Backbone	MobileNetV3	✓	✓	X
Backbone	ResNet50	✓	✓	✓
Backbone	ViT-Base	✓	✓	✓
Backbone	YOLOv8m-cls	✓	✓	✓
Object Detection	LPRNet	✓	✓	✓
Object Detection	YOLOv11m	✓	✓	✓
Object Detection	YOLOv12m	✓	✓	✓
Object Detection	YOLOv26	✓	✓	✓
Object Detection	YOLOv10m	✓	✓	✓
Object Detection	YOLOv3	✓	✓	✓
Object Detection	YOLOv5m-face	✓	✓	✓
Object Detection	YOLOv5s	✓	✓	✓
Object Detection	YOLOv7	✓	✓	✓
Object Detection	YOLOv8m	✓	✓	✓
Object Detection	YOLOv9m	✓	✓	✓
Object Detection	YOLOX	✓	✓	✓
Embedding	BGE	✓	✓	✓
Embedding	GTE-Qwen2-1.5B	✓	✓	✓
Embedding	Qwen3 Embedding 4B	✓	✓	✓
Pose Estimation	YOLOv8m-pose	✓	✓	✓
LLM	CoPaw-Flash-9B	✓	✓	✓
LLM	CPM-9g-8B	✓	✓	X
LLM	DeepSeek-8B	✓	✓	✓
LLM	GLM4.7-flash	✓	✓	X
LLM	GPT-OSS-20B-A3B	✓	✓	✓
LLM	Hunyuan-80B-A13B	✓	✓	X
LLM	Kimi-VL-16B-A3B	✓	✓	X
LLM	Qwen2.5-7B	✓	✓	✓
LLM	Qwen3-14B	✓	✓	✓
LLM	Qwen3-30B-A3B	✓	✓	✓
LLM	Qwen3-32B	✓	✓	✓
LLM	Qwen3-8B	✓	✓	✓
LLM	Qwen3.5-9B/4B/2B	✓	✓	✓	9b支持MTP演示，MTP暂不支持用户量化
LLM	Qwen3.5/3.6-27B	✓	✓	✓	支持MTP演示，MTP暂不支持用户量化
LLM	Qwen3.5/3.6-35B-A3B	✓	✓	✓
MLLM	MiniCPM-O-2_6	✓	✓	✓
MLLM	Qwen3-Omni	✓	✓	X
OCR	GLM-OCR	✓	✓	✓
OCR	PP-OCRv3	✓	✓	✓
OCR	PP-OCRv5	✓	✓	X
Segmentation	YOLOv8m-seg	✓	✓	✓
TTS	CosyVoice3-0.5B	✓	✓	✓
TTS	F5-TTS	✓	✓	X
VLA	OpenVLA	✓	✓	X
VLA	Pi0.5	✓	✓	X
VLA	SmolVLA	✓	✓	X
VLA	SpiritV1.5	✓	✓	X
VLA	x-VLA	✓	✓	X
VLM	Gemma4-26B-A4B	✓	✓	✓
VLM	Grounding-DINO	✓	✓	X
VLM	Qwen2.5-VL-7B	✓	✓	✓
VLM	Qwen3-VL-2B/4B/8B	✓	✓	✓
VLM	Qwen3-VL-30B-A3B	✓	✓	✓

3.2. 模型推理能力适配表

说明：

• 表中 “支持1芯”、“支持2芯”、 “支持4芯” 表示软件层面适配的算力规格，即后摩M50芯片数，而非硬件产品物理固有的芯片搭载数量。

• ✓ ：表示支持。

• X ：表示不支持。

• — ：表示当前版本未完成适配验证。

表 3.2 模型推理能力适配表

模型类型	模型名	支持1芯	支持2芯	支持4芯	支持多batch
ASR	FireRedASR	✓	—	—	—
ASR	GLM-ASR-Nano-2512	✓	—	—	—
ASR	Qwen3-ASR	✓	—	—	—
ASR	Whisper-large-v3-turbo-0.8B	✓	—	—	—
ASR	Whisper-medium	✓	—	—	—
Autonomous Driving	YOLOP	✓	—	—	✓
Backbone	EfficientNet	✓	—	—	✓
Backbone	MobileNetV2	✓	—	—	✓
Backbone	MobileNetV3	✓	—	—	✓
Backbone	ResNet50	✓	—	—	✓
Backbone	ViT-Base	✓	—	—	✓
Backbone	YOLOv8m-cls	✓	—	—	✓
Object Detection	LPRNet	✓	—	—	✓
Object Detection	YOLOv11m	✓	—	—	✓
Object Detection	YOLOv12m	✓	—	—	✓
Object Detection	YOLOv26	✓	—	—	✓
Object Detection	YOLOv10m	✓	—	—	✓
Object Detection	YOLOv3	✓	—	—	✓
Object Detection	YOLOv5m-face	✓	—	—	✓
Object Detection	YOLOv5s	✓	—	—	✓
Object Detection	YOLOv7	✓	—	—	✓
Object Detection	YOLOv8m	✓	—	—	✓
Object Detection	YOLOv9m	✓	—	—	✓
Object Detection	YOLOX	✓	—	—	—
Embedding	BGE	✓	—	—	—
Embedding	GTE-Qwen2-1.5B	✓	—	—	—
Embedding	Qwen3 Embedding 4B	✓	—	—	—
Pose Estimation	YOLOv8m-pose	✓	—	—	✓
LLM	CoPaw-Flash-9B	✓	—	—	—
LLM	CPM-9g-8B	✓	—	—	—
LLM	DeepSeek-8B	✓	—	—	✓
LLM	GLM4.7-flash	✓	—	—	—
LLM	GPT-OSS-20B-A3B	✓	✓	—	—
LLM	Hunyuan-80B-A13B	X	✓	—	—
LLM	Kimi-VL-16B-A3B	✓	—	—	—
LLM	Qwen2.5-7B	✓	—	—	✓
LLM	Qwen3-14B	✓	✓	—	—
LLM	Qwen3-30B-A3B	✓	✓	—	—
LLM	Qwen3-32B	X	✓	—	—
LLM	Qwen3-8B	✓	—	—	✓
LLM	Qwen3.5-9B/4B/2B	✓	—	—	X
LLM	Qwen3.5/3.6-27B	✓	✓	✓	X
LLM	Qwen3.5/3.6-35B-A3B	✓	✓	—	X
MLLM	MiniCPM-O-2_6	✓	—	—	—
MLLM	Qwen3-Omni	✓	—	—	—
OCR	GLM-OCR	✓	—	—	—
OCR	PP-OCRv3	✓	—	—	—
OCR	PP-OCRv5	✓	—	—	—
Segmentation	YOLOv8m-seg	✓	—	—	✓
TTS	CosyVoice3-0.5B	✓	—	—	—
TTS	F5-TTS	✓	—	—	—
VLA	OpenVLA	✓	—	—	—
VLA	Pi0.5	✓	—	—	—
VLA	SmolVLA	✓	—	—	—
VLA	SpiritV1.5	✓	—	—	—
VLA	x-VLA	✓	—	—	—
VLM	Gemma4-26B-A4B	✓	✓	—	—
VLM	Grounding-DINO	✓	—	—	—
VLM	Qwen2.5-VL-7B	✓	—	—	—
VLM	Qwen3-VL-2B/4B/8B	✓	—	—	—
VLM	Qwen3-VL-30B-A3B	✓	✓	—	—