Add DeepSeek-V4-Pro, Qwen3.5, MiniMax-M3 (ATOM + single-node XGMI), and GLM-5.2 inference recipes (scripts/) - #173

Open

raviguptaamd wants to merge 12 commits into

ROCm:developfrom

raviguptaamd:add-deepseekv4-recipe

Commits on Jun 29, 2026

Add DeepSeek-V4-Pro disaggregated P/D inference recipe (scripts/DeepSeekV4)
raviguptaamd
committed

Commits on Jul 8, 2026

Add Qwen3.5-35B-A3B aggregated inference recipe (scripts/Qwen3.5)

raviguptaamd
and
claude
committed

Commits on Jul 9, 2026

Qwen3.5: add lib/ files (force-add past .gitignore lib/ rule)

raviguptaamd
and
claude
committed
Qwen3.5 README: add highlighted 4-step Quickstart for the default config

raviguptaamd
and
claude
committed

Commits on Jul 16, 2026

Add MiniMax-M3 MXFP4 disaggregated P/D inference recipe (scripts/MiniMax-M3)

Ravi Gupta
and
claude
committed
MiniMax-M3: vendor InferenceX ATOM MXFP4 disagg code verbatim

Ravi Gupta
and
claude
committed
MiniMax-M3: make ATOM MXFP4 disagg recipe runnable standalone in MAD

Ravi Gupta
and
claude
committed
MiniMax-M3: trim README to plain runnable doc
Ravi Gupta
committed

Commits on Jul 23, 2026

Add MiniMax-M3 single-node P/D over XGMI recipe (scripts/MiniMax-M3-single-pd)

Ravi Gupta
and
claude
committed
MiniMax-M3-single-pd: fix docker volume paths (absolute LOG, mount bench result-dir)

Ravi Gupta
and
claude
committed
MiniMax-M3-single-pd: retry smoke inference (engine warmup after registration)

Ravi Gupta
and
claude
committed

Commits on Jul 27, 2026

Add GLM-5.2 disaggregated P/D recipe (scripts/GLM-5.2-disagg)
Ravi Gupta
committed