prismml

HIP/ROCm fork optimized for AMD RDNA2 (gfx1030) with PrismML Q1_0_G128 1-bit quant support, RotorQuant, TurboQuant, EAGLE3 and P-EAGLE speculative decoding, and full Wave32 kernel optimizations.

AMD ROCm (gfx1030) inference fork with RotorQuant/TurboQuant KV compression, PHANTOM-X zero-copy draft speculation, EAGLE3 speculative decoding, 12 RDNA2 crash fixes, and PrismML Bonsai Q1_0_G128 1-bit GGUF support.

Precompiled PrismML/llama.cpp backend for LM Studio on Windows. Run PrismML Bonsai 1-bit/ternary GGUF models natively in LM Studio with NVIDIA CUDA. No compiling required.

HTTP proxy serving PrismML Bonsai 1.7B (Q1_0 GGUF) through an Ollama-compatible API via llama-server.

Navigable Degeneracy in the Roots of 1-Bit Language Models

Windows x64 installer for Bonsai 2 / PrismML GGUF models in LM Studio. Isolates PrismML server binaries, preserves native libraries, verifies downloads, and backs up runtime manifests.

One-click Windows setup & launcher for PrismML Bonsai-demo — pick a model, everything else is automated.

GGUFly — Interactive TUI launcher and model manager for GGUF models with llama.cpp, PrismML/Bonsai, and Mirai S runtimes. Built on Omarchy / Arch Linux.

Ternary-Bonsai-2-27B (PTQ1_0 QAT) on 8GB - PrismML b10709 A/B (+15% PP/+24% TG), quality gate, production config

Local chat app for running Bonsai GGUF models with streaming responses, built-in Prism runtime management, model switching, runtime diagnostics, and saved conversation history.

Restart-proof KV cache for llama.cpp: 40K-token TTFT 155s→6s after reboot. llama-server 重启/关机后自动恢复 prompt 前缀缓存，不重算 prefill。

Terminal workspace for chatting with Bonsai AI models locally using the PrismML custom runtime.