[AMD] [AGENTX] MiniMax-M3 FP4 MTP on MI355X - #2487
Conversation
|
Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase For PR verification, add the PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. See GitHub's docs on re-running failed jobs 感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 如需进行 PR 验证，请为此 PR 添加 PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档 |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
Signed-off-by: Sirra <asirra@amd.com>
…g Search Space & KV Backend Signed-off-by: Sirra <asirra@amd.com>
…R Id to Perf Change Log. Signed-off-by: Sirra <asirra@amd.com>
…RAFT Model & Details. Signed-off-by: Sirra <asirra@amd.com>
…eptance to the golden AL Throughput replay now runs with rejection_sample_method=synthetic and synthetic_acceptance_length=3.35, the committed minimax-m3 EAGLE3 golden AL for thinking_on at num_speculative_tokens=5 (golden_al_distribution/minimaxm3_eagle3.yaml), as required by the AgentX fairness guidelines for agentic speculative-decoding submissions. EVAL_ONLY accuracy runs keep real target verification -- synthetic acceptance bypasses verification and corrupts the eval score. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
f7d82d7 to
cebeb27
Compare
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31069894361 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31069977402 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31071015341 |
Signed-off-by: Sirra <asirra@amd.com>
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31078025344 |
Signed-off-by: Sirra <asirra@amd.com>
Signed-off-by: Sirra <asirra@amd.com>
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31078326932 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31078666569 |
Signed-off-by: Sirra <asirra@amd.com>
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31079152471 |
Signed-off-by: Sirra <asirra@amd.com>
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31081873415 |
Restore EAGLE3 with the committed golden acceptance length, add TP4 and TP2 GPU-resident coverage, and add TP-sharded LMCache MP offload using the official ROCm wheel. Require vLLM Prometheus artifacts and preserve a strict no-exception benchmark path.\n\n恢复使用已提交黄金接受长度的 EAGLE3，新增 TP4/TP2 GPU 常驻搜索空间，并通过官方 ROCm wheel 增加按 TP 分片的 LMCache MP 卸载。强制生成 vLLM Prometheus 指标，并保持无例外的严格基准路径。
cquil11
added
the
agentx-fast
显式安装 LMCache MP 所需的 sortedcontainers 运行时依赖，同时保留 vLLM 镜像中经过验证的 torch/ROCm 软件栈。
…m3-tuning # Conflicts: # perf-changelog.yaml
cquil11
added
full-sweep-enabled
and removed
agentx-fast
cquil11
changed the title
…m3-tuning # Conflicts: # perf-changelog.yaml
4 participants
Add MiniMax-M3 FP4 MI355X Agentic Support with MTP
Adds a new single-node agentic-coding benchmark recipe for MiniMax-M3 (FP4) on MI355X using vLLM, with EAGLE3 MTP speculative decoding and DRAM-backed KV cache offloading (vllm-simple).
Changes
minimaxm3-fp4-mi355x-vllm-agentic-mtp(imagevllm-openai-rocm:nightly-cb8104839c141609d99f1254459ef3a4f1bd4263,tp=4, DRAM KV offloading viavllm-simple,spec-decoding: mtp).rejection_sample_method: syntheticwithsynthetic_acceptance_length: 3.35, the committed MiniMax-M3 EAGLE3 golden AL forthinking_onatnum_speculative_tokens=5(golden_al_distribution/minimaxm3_eagle3.yaml), per the AgentX fairness guidelines.EVAL_ONLYaccuracy runs keep real target verification, since synthetic acceptance bypasses verification and corrupts the eval score.Status
Marked [WIP].