[ROCm] [Release] Update ROCm variant from `rocm700` to `rocm721`#80

Open

tjtanaa wants to merge 16 commits into

EmbeddedLLM:mainfrom

tjtanaa:rocm721update

Commits on Mar 27, 2026

change the docker hub upload to nightly
tjtanaa
committed
Remove need for explicit `\n` in docstring lists for `--help` formatting (#38350)

hmellor
and
gemini-code-assist[bot]
authored
{ROCm]: gpt-oss fusion/padding fixes (#38043)

Rohan138
and
AndreasKaratzas
authored
[ROCm] Fix GPT-OSS import for triton 3.6 (#37453)
gshtras
authored
Add short flag `-sc` for `--speculative-config` argument (#38380)

mgoin
and
claude
authored
[QeRL] Compose online quantization with quantized reloading (#38032)
kylesayrs
authored
[CI] Skip failing test (#38369)
NickLucche
authored
[Model Runner V2] Rebuild attention metadata before eagle decode full… (#38311)
TheEpicDolphin
authored
[NVIDIA] Fix DGX Spark logic (#38126)

authored
[ROCm][CI/Build] ROCm 7.2.1 release version; torch 2.10; triton 3.6 (#38252)
gshtras
authored
[ROCm][Documentation] update quickstart and installation to include rocm nightly docker tips (#38367)

hongxiayang
and
Hongxia Yang
authored
Add nvidia h800 moe config (#31201)
lengrongfu
authored
[Perf] Use torch compile to fuse pack topk in trtllm moe (#37695)
wzhao18
authored
[Bugfix]fix output Nan/Inf in marlin if dtype=float16 (#33972)
ir1ka
authored

Commits on Mar 28, 2026