AXera Models Research

This is the home for Axera's npu model(axmodel) and npu's tools (Pulsar2). We released(such as):

• MiniCPM4 : MiniCPM4-0.5B
• Qwen3.5 : Qwen3.5-0.8B, Qwen3.5-2B, Qwen3.5-4B
• Qwen3 : Qwen3-0.6B, Qwen3-1.7B, Qwen3-4B
• Qwen2.5 : Qwen2.5-0.5B, Qwen2.5-1.5B, Qwen2.5-3B, Qwen2.5-7B
• DeepSeek : DeepSeek-R1-Distill-Qwen-1.5B, DeepSeek-R1-Distill-Qwen-7B
• HuggingFaceTB : SmolLM, SmolVLM, SmolVLM2
• Multimodal Models : CLIP, MobileCLIP2, JinaCLIP, StableDiffusion, Qwen3-VL-2B/4B, InternelVL3_5-1B/2B, FastVLM-0.5B/1.5B, Qwen2.5-VL-3B/7B, Janus-Pro-1B, MiniCPM4-V
• Vision Models : Ultralytics, Depth-Anything-V2, MixFormerV2, LivePortrait, Real-ESRGAN
• Audio Models : Whisper, SenseVoice, ZipFormer, CosyVoice2, MeloTTS, FireRed-AED, SileroVAD, Kokoro
• YOLO-Fun : Detection Models

Solution

• Frigate NVR : AI NVR solution, support AX650 and AXCL
• Immich : High performance self-hosted photo and video management solution

Tools

• Pulsar2 : The NPU Toolchain for AX650/AX8850, AX630C/AX620Q, AX615, AX637
• AXCL：The driver install package for AX650/AX8850
• PPQ-XS : The NPU Toolchain for AX520/AX513
• AX650-SDK : The Full SDK for AX650/AX8850

Other

• AXERA on GitHub