Kaden-Schutt/hipfire

gfx1151 (Strix Halo) speed-baseline + perf bench (help wanted) about 2 hours ago

enhancement help wanted kernel-perf

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

Lloyd-MQ3 ship gates: K4-unroll perf + coherence eyeball 2 months ago

help wanted kernel-perf research

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

[Path D] Stale-context overlap pipelining (1-2 wk eng, +5-15% projected) 2 months ago

enhancement help wanted roadmap speculative-decode

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

RDNA4 native ISA: FP8/INT4 WMMA + SWMMAC + R9700 CU fix on gfx1201 2 months ago

help wanted kernel-perf rdna-only research

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

Multi-GPU support roadmap (PP first, TP follow-up) 2 months ago

enhancement help wanted

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

hipfire chat — interactive TUI chat (Ollama-style scriptable surface stays) 2 months ago

enhancement help wanted

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

Track auto-MMQ regression on tool-call output (gfx1151) — reverted on master 2 months ago

help wanted kernel-perf

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

Prefill scaling regression vs llama.cpp at pp≥512 on 9B+ (help wanted) 2 months ago

enhancement help wanted kernel-perf

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

DFlash thinking-attractor bug on A3B drafts (qwen3.5/3.6 35b-a3b) — block loop in <think> at long budgets 2 months ago

help wanted speculative-decode

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

Port sliding-window FA from Lucebox PR #26 (long-context decode lever, 3.48× projected) 3 months ago

enhancement help wanted kernel-perf

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

[research] Fix DDTree on gfx1100: RoPE phase-delta skew at FA layers 3 months ago

help wanted roadmap speculative-decode rdna-only research

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

gfx908 / MI100 (CDNA1): port MFMA prefill kernels (4 kernels + channel-tests) 3 months ago

help wanted kernel-perf

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

gfx12 WMMA: tune 9B prefill (multi-row, K-tile, s_prefetch, launch_bounds) 3 months ago

help wanted kernel-perf

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

Vulkan backend MVP: 4B Qwen AR decode on gfx1100 + gfx1010 (#84) 3 months ago

enhancement help wanted roadmap

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

Phase 3 prompt-shape: generalized rare-token rewriting beyond \n{3,} 3 months ago

enhancement good first issue roadmap

Kaden-Schutt/hipfire

474

RDNA-native LLM inference engine in Rust.

Rust

#amd-gpu#gpu-computing#hip#llm-inference#machine-learning#quantization#rdna#rocm#rust

Open Issues Need Help