Repository navigation
[AgentX] add Muse Spark Open H100 concurrency 1–64 sweep / [AgentX] 新增 Muse Spark Open H100 并发 1–64 扫描 - #4149
Merged
Merged
Conversation
Contributor
|
Thanks for the contribution!
中文感谢你的贡献！
|
edwingao28
added
the
full-sweep-fail-fast
Contributor
|
View unofficial run (performance): https://inferencex.semianalysis.com/inference?unofficialRun=38878624849 View unofficial run (accuracy): https://inferencex.semianalysis.com/evaluation?unofficialRun=38878624849 |
中文：同步主分支并保留已测提交和当前配置。
扩展 H100 并发扫描至 1–64，并同步主分支。
同步 B300 主分支集成并保留 H100 完整并发扫描。
同步 GB300 主分支集成并保留 H100 完整并发扫描。
edwingao28
changed the title
限制 H100 解码图批量并保留完整并发扫描。
同步主分支并保留 H100 扫描矩阵及解码图修复。
…gentx # Conflicts: # inferencex-e2e/perf-changelog.yaml
Oseltamivir
marked this pull request as ready for review
October 16, 2026 07:12
Oseltamivir
requested review from
Ankur-singh,
RohitNagraj,
kedarpotdar-nv,
nvpohanh and
wzhao18
as code owners
October 16, 2026 07:12
Oseltamivir
added
non-canary-full-sweep-enabled
Contributor
|
Claude finished @Oseltamivir's task in 34s —— View job LGTM - no blocking issues found. What I checked
|
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
I reviewed this PR and didn't find any bugs. The new recipe and master entry are in CODEOWNER-owned NVIDIA paths, and the description says the C64 graph-allocation fix hasn't passed a sweep yet, so a human look is still worthwhile.
What was reviewed:
- Recipe against master entry: srt-recipe-dir, srt-recipe, and TP8/EP8 match. The 9-point zip lists (max-running-requests, cuda-graph-max-bs-decode, CONC) line up with conc-list 1..64.
- Compared the recipe with the sibling B300 Muse Spark recipe. The draft checkpoint and DFlash2 settings are unchanged, with no draft-dtype overrides and no hard-coded acceptance lengths.
- The runners.yaml models.entries row is in sorted order. The perf-changelog block is appended at the tail as a single English entry for the new key.
Extended reasoning...
The change adds one new H100 SGLang AgentX srt-slurm recipe, a matching nvidia-master.yaml entry with an image pinned by digest, one runners.yaml model-path row, and an appended perf-changelog entry. It touches no security-sensitive surface. It is configuration only, but .github/CODEOWNERS assigns both nvidia-master.yaml and the h100-* recipe tree to specific NVIDIA owners. The PR description also says the C64 graph-memory fix is still waiting on a sweep, so a human should confirm the results before merging.
c48 started with 0.29 GB free per GPU and OOMed in the vision encoder's all-gather on the first image prefill.
启用 eager draft sampler，保留解码图批量上限 8 和完整并发扫描。
Batches above the batch-8 decode graphs run eager and OOM in fused MoE with ~1 GB free per GPU (c48, run 38870885984). Drop c24-c64.
Collaborator
|
/use 38878624849 |
Contributor
|
@Oseltamivir staged run 38878624849. This run remains available across future |
…e Spark Open H100 添加 HiCache 卸载 (#4181) * feat: add HiCache offload arm to H100 Muse Spark at c2-c8 * chore: link H100 HiCache entry to PR 4181
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Add H100 TP8/EP8 Muse Spark Open NVFP4 with Marlin W4A16. Full sweep: C1/2/4/8/16/24/32/48/64 plus C64 natural GSM8K. Cap decode graphs at 8; use eager DFlash sampling above graph capacity.
Testing: Prior C48/C64 startup hit NCCL OOM. Replacement sweep at
266c72c1fpassed setup; GPU qualification remains pending.Blocker: Full results, cookbook, verified reuse and human review.
中文
新增 H100 TP8/EP8 Muse Spark Open NVFP4，使用 Marlin W4A16。完整扫描覆盖 C1/2/4/8/16/24/32/48/64，以及 C64 自然验证 GSM8K。解码图批量上限为 8；超过图容量时使用 eager DFlash 采样。
**测试：**此前 C48/C64 启动时出现 NCCL OOM。
266c72c1f的替代扫描已通过 setup，GPU 验收仍待完成，链接见上方。**阻塞：**完整结果、上游 cookbook、验证后的结果复用及人工审查。
AI model disclosure
PR checklist and change type
Type of Change
Checklist
python -m infx.matrix.changed --base origin/mainlists affected config keys, I have appended one new entry covering them to the physical end ofinferencex-e2e/perf-changelog.yaml; if it lists none, I have added none. I have not edited historical entriesinferencex-e2e/perf-changelog.yamlentry, it carries exactly one primary sweep label (a maintainer applies it on fork PRs):full-sweep-fail-fast(recommended),full-sweep-enabled, ornon-canary-full-sweep-enabled. Optional modifiersall-evals,evals-only, andagentx-fastrequire a primary label; the last two block reuse while applied.OWNER/MEMBER/COLLABORATOR) has commented/use <run_id>(or the legacy/reuse-sweep-run) on this PR. Do this only once there is a final full sweep that is all green with evals passing, since after this comment the primary sweep label will no longer automatically kick off new sweeps. Remove and re-add the primary sweep label to force a new sweep.