~/bookmarks

William's Bookmark Library

/**/

GitHub - julianmb/halofpx: High-performance Lemonade alternative for AMD Strix Halo & Radeon — latest MoE models (Ling-3.0-Flash, Qwen 3.8 Flash Next, Ornith 1.5), bleeding-edge RDNA 3.5 Wave64/ROCmFP4 kernels, and silicon-tuned profiles.

github.comSaved October 1, 20269 min

Preview of GitHub - julianmb/halofpx: High-performance Lemonade alternative for AMD Strix Halo & Radeon — latest MoE models (Ling-3.0-Flash, Qwen 3.8 Flash Next, Ornith 1.5), bleeding-edge RDNA 3.5 Wave64/ROCmFP4 kernels, and silicon-tuned profiles.

julianmb · via GitHub

Summary

HaloFPX is an OpenAI-compatible model server for AMD Strix Halo APUs and Radeon GPUs with full Lemonade CLI and cache parity.

It uses Mesa RADV Wave64 cooperative matrices and ROCmFP4 quantization on RDNA 3.5, running models including Ling-3.0-Flash (124B), Qwen 3.8 Flash Next (125B), and Ornith 1.5 (35B). Validated at 262K context with TurboQuant KV-cache across 4 slots (524K total tokens) in unified memory.

Topics

AMD Strix Halo RDNA 3.5 Local LLM Inference MoE Models ROCm

View on GitHub All Bookmarks

William's Bookmark Library

/**/

GitHub - julianmb/halofpx: High-performance Lemonade alternative for AMD Strix Halo & Radeon — latest MoE models (Ling-3.0-Flash, Qwen 3.8 Flash Next, Ornith 1.5), bleeding-edge RDNA 3.5 Wave64/ROCmFP4 kernels, and silicon-tuned profiles.

github.comSaved October 1, 20269 min

julianmb · via GitHub

Summary

HaloFPX is an OpenAI-compatible model server for AMD Strix Halo APUs and Radeon GPUs with full Lemonade CLI and cache parity.

It uses Mesa RADV Wave64 cooperative matrices and ROCmFP4 quantization on RDNA 3.5, running models including Ling-3.0-Flash (124B), Qwen 3.8 Flash Next (125B), and Ornith 1.5 (35B). Validated at 262K context with TurboQuant KV-cache across 4 slots (524K total tokens) in unified memory.

Topics

AMD Strix Halo RDNA 3.5 Local LLM Inference MoE Models ROCm

View on GitHub All Bookmarks

~/bookmarks

GitHub - julianmb/halofpx: High-performance Lemonade alternative for AMD Strix Halo & Radeon — latest MoE models (Ling-3.0-Flash, Qwen 3.8 Flash Next, Ornith 1.5), bleeding-edge RDNA 3.5 Wave64/ROCmFP4 kernels, and silicon-tuned profiles.

Summary

Topics

Discover Similar Content

GitHub - julianmb/q38rocm: Qwen 3.8 27B ROCmFP4 on AMD Strix Halo (Ryzen AI Max+ 395). Up to 36 tok/s via MTP Speculation, TurboQuant & Mesa RADV Wave64.

GitHub - kyuz0/strix-halo-ds4-toolbox

vmlinux/Muse-Glimmer-30B-ROCmFPX-GGUF · Hugging Face

Discover Similar Content

GitHub - julianmb/halofpx: High-performance Lemonade alternative for AMD Strix Halo & Radeon — latest MoE models (Ling-3.0-Flash, Qwen 3.8 Flash Next, Ornith 1.5), bleeding-edge RDNA 3.5 Wave64/ROCmFP4 kernels, and silicon-tuned profiles.

Summary

Topics

Discover Similar Content

GitHub - julianmb/q38rocm: Qwen 3.8 27B ROCmFP4 on AMD Strix Halo (Ryzen AI Max+ 395). Up to 36 tok/s via MTP Speculation, TurboQuant & Mesa RADV Wave64.

GitHub - kyuz0/strix-halo-ds4-toolbox

vmlinux/Muse-Glimmer-30B-ROCmFPX-GGUF · Hugging Face