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.
-
Updated
Jun 23, 2026 - C++
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.
llama.cpp-PrismML-Fork-LmStudio - Precompiled PrismML/llama.cpp backend for LM Studio on Windows for GPU and CPU without errors or Garbage Output. Run PrismML Bonsai 1 and 2 1-bit/ternary GGUF models natively in LM Studio with NVIDIA CUDA and CPU. No compiling required.
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.
Ternary Bonsai 2: New 0.4GB Vision AI With 310 Tokens/Sec Speed! - 100% private visual document scanner and PII redactor running locally.
Ternary Bonsai 2 9B on a GTX 1080 (sm_61): +51% decode with bit-identical output (PrismML llama.cpp fork, PQ2_0 + MTP)
Navigable Degeneracy in the Roots of 1-Bit Language Models
基于 llama.cpp b-10991 的 Windows x86-64 推理 / 压缩 / 训练引擎:为 Bonsai Q1_0 提供专用 CPU 内核,同时提供面向更多模型的 Vulkan 加速、自动设备策略、无损压缩与 QLoRA 训练。
One-click Windows setup & launcher for PrismML Bonsai-demo — pick a model, everything else is automated.
Run Ternary Bonsai 2 27B on free Google Colab (Tesla T4) at ~30 tok/s. Zero-OOM pipeline with full CUDA acceleration and OpenAI-compatible API via Cloudflare.
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。
Local deployment of Ternary Bonsai 2 27B (PrismML 1.75-bit ternary quantized Qwen3.8 27B) on an RTX 4060 laptop — setup, benchmarks and notes
A local, air-gapped multi-agent coding harness: sequential agent roles on local models, tools in a Podman sandbox, and nothing red ever committed.
Byte-level reverse engineering of PrismML Ternary Bonsai 2 27B: 1.58-bit quantization, 42/128 zero mask, Hadamard rotation, GGUF formats, and fact-checked benchmarks.
To associate your repository with the prismml topic, visit your repo's landing page and select "manage topics."