b4rtaz/distributed-llama

Distributed LLM inference. Connect home devices into a powerful cluster to accelerate LLM inference. More devices means faster inference.

55
/ 100
Established

Implements tensor parallelism across a root-worker architecture where the root node loads and distributes model slices over Ethernet to worker nodes, with synchronized state updates using configurable float precision (q80/f32). Supports CPU inference on ARM and x86_64 with AVX2 optimization, plus experimental Vulkan GPU acceleration, enabling models like Llama 3.3 70B and Qwen 3 MoE across commodity hardware with quantized weights (Q40).

2,856 stars.

No Package No Dependents
Maintenance 10 / 25
Adoption 10 / 25
Maturity 16 / 25
Community 19 / 25

How are scores calculated?

Stars

2,856

Forks

215

Language

C++

License

MIT

Last pushed

Feb 10, 2026

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/transformers/b4rtaz/distributed-llama"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.