Qwen 3.8 27b q8 unsloth running in Hermes Agent, like it owns the place...
With a whole lot of stuff loaded... vanilla most things + the five custom-mcp's in the install... q8 gets ~30 tps on a 48gb (modded) RTX 4090 😅 Should i throw "something gnarly at it" and just let it
Qwen 3.8 is a 2.4 TRILLION parameter model built to work for DAYS without you.
Alibaba showed it running multi-day coding projects. Writing, testing, fixing. Zero human clicks. 🤯 The specs: → 2.4 trillion
Qwen 3.8 27b q5_k_m tiny grammar gbnf file (see 2nd post) and mtp 3 on 5090, 110 t/s
Qwen 3.8 - 27B MLX-Serve 4bit model, made the best version of the famous Pagoda test so far for me.
Using pidotdev as the coding agent, and MLX-Serve as the backend.
You have reached the end of the archive
All of qwen38