mw00 / project-maya
git clone https://github.com/mw00/project-maya.git
v1.1.19 · 12d ago
Local inference engine and server for running GLM-5.3-Flash and DeepSeek-V4.1-Flash on your own NVIDIA, AMD, or Windows GPUs with an OpenAI/Anthropic-compatible API.
Why you might care
Targets 300B-750B mixture-of-experts models that normally need a datacenter, by tiering experts across GPU, RAM, and NVMe and tuning to your hardware. Offers its own quantized GLM builds, an OpenAI- and Anthropic-compatible API with tool calls, and parallel requests; nothing leaves the machine.
+45
stars, 24h · #913
+477
stars, 7 days · #580
+2.4k
stars, 30 days · #428
2.4k
★ total · +25% this week
326
forks · 139 open issues
Star history (as recorded here)
LanguageCuda
LicenseMIT
Created2026-10-06 (26d ago)
Last pushtoday
Watchers6
OwnerUser
GitHub topics
cudadeepseekdeepseek-v4glmglm-5llmllm-inferencelocal-llmminimaxminimax-m3mixture-of-expertsmoeopenai-apiquantizationrocmself-hosted
Links
- GitHub: github.com/mw00/project-maya
- Homepage: https://huggingface.co/peasantsmith
- Releases · Issues · Hacker News mentions
README
Loading README…
Blurb generated by claude-haiku-5-5 7d ago. Star deltas marked ~ are extrapolated from partial history.