qwen3-6

Use PEFT or Full-parameter to CPT/SFT/DPO/GRPO 600+ LLMs (Qwen3.6, DeepSeek-V4, GLM-5.1, InternLM3, Llama4, ...) and 300+ MLLMs (Qwen3-VL, Qwen3-Omni, InternVL3.5, Ovis2.5, GLM4.5v, Gemma4, Llava, Phi4, ...) (AAAI 2025).

FlashRT is a high-performance realtime inference engine for small-batch, latency-sensitive AI workloads. The flagship integration is production VLA control for Pi0, Pi0.5, GROOT N1.6, and Pi0-FAST. Also support llm e.g, qwen3.6-27B

Docker Compose for llama.cpp GGUF servers on AMD Strix Halo: Qwen, Gemma, and Laguna packages (abliterated and quantized), stock Vulkan plus ROCmFP4/MTP and ROCmFPX, parallel slots, with prefill/decode and quality metrics measured on this rig.

Local AI assistant backend — FastAPI, autonomous web search, RAG, and streaming. Runs fully offline on macOS

Native Swift + Metal local LLM inference for Qwen 3.6 35B-A3B on Apple Silicon. Streams active MoE experts from SSD for ~5 GiB Metal allocation; reads MLX Q4 Safetensors and includes a macOS chat app and benchmarks.

Reproducible single-GPU Qwen3.6-35B-A3B W8A8 inference tuning for Hygon K100AI (gfx928) on vLLM 0.18.1.

A native macOS AI assistant that runs large language models fully on-device on Apple Silicon via Apple MLX (mlx-swift-lm). No server, no account, no cloud — your conversations and files never leave your Mac.

SahurHub is a deliberately useless task reminder that steals to-dos from ordinary conversation and nags through an animated 3D character on a Raspberry Pi, built on Bun, Three.js and Qwen.

Reproducible Qwen3.6-35B-A3B inference optimization for MLX-LM on Apple M4 Pro, with guarded Metal kernels and full benchmark evidence.