fix: pick best AMD GPU (dGPU > iGPU) in get_rocm_arch() instead of first available - #2459

Closed

bong-water-water-bong wants to merge 14 commits into

lemonade-sdk:mainfrom

bong-water-water-bong:fix/rocm-best-gpu

Commits on Jun 27, 2026

fix: make get_rocm_arch() prefer the best AMD GPU (dGPU > iGPU, more VRAM > less)
bong-water-water-bong
committed
caac78a
fix: remove unused rocm_arch_score param, align comment, add selection logging
bong-water-water-bong
committed
654840f
fix(vllm): set VLLM_TARGET_DEVICE=rocm to prevent dual-platform crash on hybrid AMD+NVIDIA laptops

bong-water-water-bong
and
claude
committed
17a0306
fix(llamacpp): hide CUDA devices from ROCm backend to fix hybrid-GPU memory crash

bong-water-water-bong
and
claude
committed
4d5fc84
fix: prevent NPU backend hangs — CURL timeout, watchdog for all backends, memory guard

bong-water-water-bong
and
claude
committed
55d5133
fix(vulkan): prevent GPU hang on Windows — flash-attn off, GPU driver cooldown, TDR warning

bong-water-water-bong
and
claude
committed
a9ea45e
fix: make model load timeout configurable via global_timeout for all backends

bong-water-water-bong
and
claude
committed
c690983
feat: add PR-Agent and Qodo Merge workflows for automated PR review

bong-water-water-bong
and
claude
committed
8008a5e

Commits on Jun 28, 2026

fix: NPU kernel safety check and download memory pressure guard (#1003)

bong-water-water-bong
and
claude
committed
1fb483b
fix: streaming inference timeout after 5 minutes on ROCm (#1792)

bong-water-water-bong
and
claude
committed
5eb43e3
fix: Vulkan VK_ERROR_DEVICE_LOST on Linux with large contexts (#2180)

bong-water-water-bong
and
claude
committed
891409c
feat: per-hardware max_loaded_models (#1705)

bong-water-water-bong
and
claude
committed
99bcd1d
fix: disable MTP speculative decoding for GLM-4 MoE models (#2451)

bong-water-water-bong
and
claude
committed
bb35f4f
fix: add Conflicts/Replaces for lemonade-server-minimal in DEB packaging

bong-water-water-bong
and
claude
committed
f6f01d4