README.md

paddlemix examples 目录下提供模型的一站式体验，包括模型推理、模型静态图部署、模型预训练，调优等能力。帮助开发者快速了解 PaddleMIX 模型能力与使用方法，降低开发门槛。

支持模型

Model	Model Size	Template
YOLO-World	640M/800M/1280M	yolo_world
audioldm2	346M/712M	audioldm2
blip2	7B	blip2
clip	2539.57M/1366.68M/986.71M/986.11M/427.62M/149.62M/151.28M	clip
coca	253.56M/638.45M/253.56M/638.45M	coca
CogVLM && CogAgent	17B	cogvlm_cogagent
eva02	6M/22M/86M/304M	eva02
evaclip	1.1B/1.3B/149M/428M/4.7B/5.0B	evaclip
groundingdino	172M/341M	groundingdino
imagebind	1.2B	imagebind
InternLM-XComposer2	7B	internlm_xcomposer2
Internvl2	1B/2B/8B/26B/40B	internvl2
janus	1.3B	janus
llava	7B/13B	llava
llava_critic	7B	llava_critic
llava_denseconnector	7B	llava_denseconnector
llava_next	0.5B/7B	llava_next_interleave
llava_onevision	0.5B/2B/7B	llava_onevision
minicpm-v-2_6	8B	minicpm_v_2_6
minigpt4	7B/13B	minigpt4
minimonkey	2B	minimonkey
qwen2_vl	2B/7B/72B	qwen2_vl
qwen_vl	7B	qwen_vl
sam	86M/307M/632M	sam
visualglm	6B	visualglm

模型能力矩阵

Model	Inference	Pretrain	SFT	LoRA	Deploy	NPU Training
YOLO-World	✅	❌	❌	❌	❌	❌
audioldm2	✅	❌	❌	❌	❌	❌
blip2	✅	✅	✅	✅	❌	❌
clip	✅	✅	❌	❌	❌	❌
coca	✅	✅	❌	❌	❌	❌
CogVLM && CogAgent	✅	❌	❌	❌	❌	❌
eva02	✅	✅	✅	❌	❌	❌
evaclip	✅	✅	❌	❌	❌	❌
groundingdino	✅	❌	🚧	❌	✅	❌
imagebind	✅	❌	❌	❌	❌	❌
InternLM-XComposer2	✅	❌	✅	❌	❌	❌
Internvl2	✅	❌	✅	❌	❌	✅
janus	✅	❌	❌	❌	❌	❌
llava	✅	✅	✅	✅	🚧	✅
llava_critic	✅	❌	❌	❌	❌	❌
llava_denseconnector	✅	❌	❌	❌	❌	❌
llava_next	✅	❌	❌	❌	❌	❌
llava_onevision	✅	❌	❌	❌	❌	❌
minicpm-v-2_6	✅	❌	❌	❌	❌	❌
minigpt4	✅	✅	✅	❌	✅	❌
minimonkey	✅	❌	✅	❌	❌	❌
qwen2_vl	✅	❌	✅	🚧	❌	❌
qwen_vl	✅	❌	✅	✅	✅	❌
sam	✅	❌	❌	❌	✅	❌
visualglm	✅	❌	✅	✅	❌	❌

• ✅: Supported
• 🚧: In Progress
• ❌: Not Supported