You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
"description": "1bit engine: model-agnostic local LLM inference on the AMD Ryzen AI MAX+ 395 (Strix Halo) inside Lemonade Server, on Linux: HRX with HIP kernels on the Radeon 8060S iGPU and our own runtime on the XDNA 2 NPU, behind an OpenAI-compatible API.",
"folders": [
"docs"
],
"excludeFolders": [
"third_party",
"tools",
"tests",
"patches"
],
"excludeFiles": [
"LICENSE",
"NOTICE"
],
"rules": [
"Serve models with `1bit serve`; it exposes an OpenAI-compatible API and runs as Lemonade's `onebit` backend.",
"Pick the device with `--device` (auto, npu, hrx, cpu); an NPU model directory runs on the NPU fast lane, and `auto` runs plain GGUF files on HRX. On the GPU the engine is pure HRX with HIP kernels (no Vulkan, ROCm backend, Metal, ZINC or CUDA): Lemonade's own backends serve those.",
"Prefer Unsloth Dynamic GGUFs: UD-Q4_K_XL is the lean default, UD-Q5_K_XL the accurate choice.",
"For one user at a time use `--mtp`; for many simultaneous requests use `--parallel`.",
"Documentation lives at https://1bit.gg; measured results are on the GitHub wiki."