rocm: Qwen3.6-35B-A3B GPU runtime — runs on Radeon 780M at ~14 tok/s, 5/5 top-1 parity#4

Open

marmyx77 wants to merge 10 commits into

andreaborio:mainfrom

marmyx77:rocm-qwen

Commits on Jul 15, 2026

rocm: ROCm 6.3.1 build shims (hipBLAS 2.x, warp-sync builtins, __syncwarp)
Marco Armellino
committed
rocm: implement the host memory snapshot for the adaptive cache planner
Marco Armellino
committed
ssd: gate the adaptive planner on snapshot capability, not backend
Marco Armellino
committed
rocm: forward-port the La Bestia hardening set
Marco Armellino
committed
build: pass the git sha as a bare -D token and stringify in ds4_build.c
Marco Armellino
committed
rocm: Qwen3.6 HIP kernels + entry points + per-op parity test
Marco Armellino
committed
test: extend qwen35 ROCm parity to the weight-reading ops
Marco Armellino
committed
rocm: wire the Qwen3.6 GPU path (resident) + top-8 MoE via two 4-expert halves
Marco Armellino
committed
docs: Qwen3.6-35B-A3B ROCm result — runs on 780M at ~14 t/s, 5/5 top-1 parity
Marco Armellino
committed
chore: drop stray .cuh.orig patch leftover, gitignore test binaries
Marco Armellino
committed