[Primus weekly report] 2026-W17#683
Merged
Merged
Conversation
Co-authored-by: wenxie-amd <wenxie-amd@users.noreply.github.com>
wenxie-amd
marked this pull request as ready for review
April 22, 2026 10:52
wenxie-amd
requested review from
Xiaoming-AMD,
limou102 and
wenxie-amd
as code owners
April 22, 2026 10:52
Contributor
There was a problem hiding this comment.
Pull request overview
Adds the autogenerated Primus weekly engineering report for ISO week 2026-W17, capturing merged PR activity plus upstream drift snapshots for Megatron-LM, torchtitan, and Primus-Turbo pins.
Changes:
- Add weekly report document for 2026-W17 with executive summary and PR table.
- Record Megatron-LM / torchtitan upstream drift status and recommendations.
- Record Primus-Turbo month-to-date pin drift status and source links.
|
|
||
| ## 2. Executive Summary | ||
|
|
||
| - **1 PR merged to `main`** in the weekly window (Mon 2026-04-20 00:00 GMT+8 → now). |
There was a problem hiding this comment.
The executive summary says the window is "Mon 2026-04-20 00:00 GMT+8 → now", but the report defines a specific end timestamp in the "Time Window" section. To avoid ambiguity when reading the report later, please replace "now" with the explicit end time (e.g., "Wed 2026-04-22 18:36 GMT+8").
Suggested change
| - **1 PR merged to `main`** in the weekly window (Mon 2026-04-20 00:00 GMT+8 → now). | |
| - **1 PR merged to `main`** in the weekly window (Mon 2026-04-20 00:00 GMT+8 → Wed 2026-04-22 18:36 GMT+8). |
Comment on lines
+5
to
+6
| - Start: Monday 2026-04-20 00:00:00 Asia/Shanghai (GMT+8) | ||
| - End: Wednesday 2026-04-22 18:36 Asia/Shanghai (GMT+8) (report generation time) |
There was a problem hiding this comment.
Time formatting is inconsistent: the start timestamp includes seconds (00:00:00) while the end timestamp does not (18:36). Consider standardizing both to the same precision (either include seconds on both or omit on both) for easier machine parsing and consistency across reports.
wenxie-amd
approved these changes
Apr 22, 2026
wenxie-amd
added a commit
that referenced
this pull request
Apr 24, 2026
## Summary Refreshes the automated Primus weekly engineering report for ISO week **2026-W17** (Mon 2026-04-20 00:00 GMT+8 → Fri 2026-04-24 09:10 GMT+8) at `docs/weekly_reports/2026-W17-primus-weekly.md`. The previous W17 report (PR #683, merged 2026-04-23 07:11 GMT+8) only covered through Wed 2026-04-22 18:36 GMT+8 and captured 1 merged PR. This refresh extends the window to the full work week, bringing the total to **6 merged PRs** and refreshing upstream drift snapshots. ## Key findings - **6 PRs merged to `main`** this week (vs 1 in the initial W17 run): - Bug Fix: **#672** (MoE overlap schedule-plan `post_attn` drift), **#671** (muon optimizer signature hardening), **#674** (recompute_layer_patches byte-identical rewrite + UT). - Performance Optimization: **#673** (parallel `pp_warmup` — bit-identical loss + iter-1 speedup UT). - CI/Infra: **#678** (JAX runner → TAS node). - Docs: **#683** (initial W17 weekly report). - **Megatron-LM drift**: pin `d3528a21` is now **361 commits behind** upstream `3034d868` (MoE router, MTP token-per-expert logging, `MambaModel` → `HybridModel` rename landed outside `megatron/core`, NVRx async compat, DDP parameter-layout refactor, NVFP4/MXFP8/TE `v2.14`). Recommendation: **plan sync**. - **torchtitan drift**: pin `5fb7cc2e` (2025-10-15) is **513 commits behind** upstream `36dd1e9d` (GraphTrainer precompile with SAC + FSDP bucketing, CooR for DeepSeek-V3, CUDA-graph annotation pass, MoE token dispatcher, Fused QKV GQAttention, FlexAttention CP, FSDP2 `fully_shard`). Recommendation: **urgent sync**. - **Primus-Turbo month-to-date drift**: none. Both `ci.yaml` (`333b68d7`) and `benchmark.yaml` (`a4488f6c`) `PRIMUS_TURBO_COMMIT` pins are unchanged vs 2026-03-30. Recommendation: **monitor**. - This week was dominated by Megatron backend hardening against upstream API drift (`post_attn` removal, `get_megatron_optimizer()` signature drift, `TransformerBlock._checkpointed_forward` closures), plus one PP-warmup perf optimization. ## Notes - Report generated automatically (cron-triggered weekly automation). No email is sent. - All upstream SHAs and commit counts are snapshots at report generation time; source references are provided in the "Source Links" section of the report. <div><a href="https://cursor.com/agents/bc-5692a988-3308-437b-8123-20bcf0390b21"><picture><source media="(prefers-color-scheme: dark)" srcset="https://cursor.com/assets/images/open-in-web-dark.png"><source media="(prefers-color-scheme: light)" srcset="https://cursor.com/assets/images/open-in-web-light.png"><img alt="Open in Web" width="114" height="28" src="https://cursor.com/assets/images/open-in-web-dark.png"></picture></a> <a href="https://cursor.com/automations/f10f6398-77ce-49c9-a984-e9fce6dd8a7a"><picture><source media="(prefers-color-scheme: dark)" srcset="https://cursor.com/assets/images/view-automation-dark.png"><source media="(prefers-color-scheme: light)" srcset="https://cursor.com/assets/images/view-automation-light.png"><img alt="View Automation" width="141" height="28" src="https://cursor.com/assets/images/view-automation-dark.png"></picture></a> </div> Co-authored-by: Cursor Agent <cursoragent@cursor.com> Co-authored-by: wenxie-amd <wenxie-amd@users.noreply.github.com>
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
![https://cursor.com/assets/images/open-in-web-dark.png"><source](https://cursor.com/assets/images/open-in-web-dark.png"><source){kind=link}
![https://cursor.com/assets/images/open-in-web-light.png"><img](https://cursor.com/assets/images/open-in-web-light.png"><img){kind=link}
![https://cursor.com/assets/images/open-in-web-dark.png"></picture></a> <a](https://cursor.com/assets/images/open-in-web-dark.png"></picture></a> <a){kind=link}
![https://cursor.com/assets/images/view-automation-dark.png"><source](https://cursor.com/assets/images/view-automation-dark.png"><source){kind=link}
![https://cursor.com/assets/images/view-automation-light.png"><img](https://cursor.com/assets/images/view-automation-light.png"><img){kind=link}
![https://cursor.com/assets/images/view-automation-dark.png"></picture></a> </div](https://cursor.com/assets/images/view-automation-dark.png"></picture></a> </div){kind=link}
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.
Summary
Adds the automated Primus weekly engineering report for ISO week 2026-W17 (Mon 2026-04-20 00:00 GMT+8 → Wed 2026-04-22 18:36 GMT+8).
Report path:
docs/weekly_reports/2026-W17-primus-weekly.mdKey findings
mainthis week — fix(megatron): adapt recompute_layer_patches to the upstream Megatron and add UT #674fix(megatron): adapt recompute_layer_patches to the upstream Megatron and add UT(category: Bug Fix).d3528a21is 344 commits behind upstreambbc6b4d1(MoE routing, NVFP4 DDP, MXFP8+FP8 DPA, Mamba→Hybrid rename, TEv2.14, FSDP fixes). Recommendation: plan sync.5fb7cc2e(2025-10-15) is 502 commits behind upstream71291346(GraphTrainer/precompile, MoE token dispatcher, Fused QKV GQAttention, FlexAttention CP, FSDP2fully_shard). Recommendation: urgent sync.ci.yaml(333b68d7) andbenchmark.yaml(a4488f6c)PRIMUS_TURBO_COMMITpins are unchanged vs 2026-03-30. Recommendation: monitor.Notes