Repository Issue Activity (beta)

#113 FP4 STRIX_LEAN + DFlash in one build — Strix Halo numbers + image request

#121 [Discussion] Complex Agentic C-coding (Cline + VS Code): Qwen3.6-27B vs. MoE Active Parameter Limits

#114 llama-rocm-7.2.4 not calling tools with PI agent

#118 ROCm 7.14 shows a 10% regression in PP for similar TG versus ROCm 7.2.4 - so what is the advantage to update?

#95 MiniMax-2.7 crash llama.cpp

#80 Using toolbox images with podman alone

#112 Ryzen AI NPU via FastFlowLM not covered

#111 Heads up on benchmarking

#107 Feat: Testing Hybrid ONNX models

#110 Missing rpc-server on llama-rocm-7.2.4

#85 Gemma 4 and updating llama.cpp ?

#106 Model evaluation request: gemma-4-12B-agentic-fable5-composer2.5-v2 on Strix Halo

#101 llama toolboxes have broken embedded webui

#103 qwopus models

#104 amd_iommu=off breaks s0i3/s2idle hardware sleep on Strix Halo laptops (0% s0ix residency) — ROG Flow Z13

#102 README.md: toolbox/distrobox command missing '--'

#97 not using a toolbox

#99 forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory)

#100 declare -A BENCH_BINS in run_benchmarks.sh is missing the rocm-7_2_4 entry

#93 Issues with MTP

#98 llama.cpp + TurboQuant KV compression on Strix Halo: setup guide & benchmark results (Qwen3-Coder-Next 80B)

#96 Whisper Integration

#91 Note: setting amd_iommu=off breaks wifi on Gmktec evo-x2 (MT7925e wifi card)

#94 [Bug] Missing libllama-server-impl.so in rocm-7.2.3 image (might be due to llama.cpp commit bb28c1f)

#82 Anyone trying the turboquant llama.cpp fork ?