Skip to content

feat(migration): sync 23 recent AIC PRs through ce2824e - #73

Merged
jasonqinzhou merged 2 commits into
mainfrom
jasonzho/sync-recent-aic-tail
Sep 1, 2026
Merged

jasonqinzhou merged 2 commits into
mainfrom
jasonzho/sync-recent-aic-tail

Conversation

@jasonqinzhou

Copy link
Copy Markdown
Contributor

Summary

  • advance the recorded AIConfigurator synchronization boundary from 095f58a51c4ca8e61b66ec108d86f223f8d559ce to current AIC main at ce2824e8abd9bef71c3b162f651e63704a0eb4c1
  • migrate the 23 first-parent AIC PRs in that tail, including version slots/data pruning, GLM-5.3, attention lanes, WideEP/MoE collectors, recommendation changes, deprecation warnings, and current performance data
  • preserve AISimulate's unified wheel/crate, active workflow, and generated ownership boundaries while adapting the upstream Cargo dependency and combined lockfile
  • fix the binary patch renderer so extended rename/copy headers are rewritten into AISimulate mirror paths
  • regenerate the perf-data reuse manifest from the final AISimulate tree and record the exact source ledger/manual-path dispositions

Source ledger

  • AIC range: 095f58a51c4ca8e61b66ec108d86f223f8d559ce..ce2824e8abd9bef71c3b162f651e63704a0eb4c1
  • AISimulate base: 0f0f4b33d61283d7c05a95289ac9da58c57e92c2
  • review units: 23 first-parent AIC PRs, listed in source order in docs/repository-history.md
  • upstream mirror envelope: 1,445 paths, including 554 binary blobs
  • both repository heads were refreshed immediately before publication and still matched the SHAs above

Intentional adaptations

  • plotext<6 was already present in the combined Python manifest/lock.
  • log = "0.4" was added to the combined aisimulate-core crate and Cargo.lock was regenerated.
  • The regenerated manifest keeps vLLM's pruned logical chunk_gated_delta_rule lane absent. The imported pin now asserts the final data state instead of reviving the stale donor lane from the final AIC merge snapshot.
  • The vLLM 0.20.1 golden allowlist now includes the migrated --gpu-memory-utilization output, which is independently covered by sweeper request tests.

Validation

  • cargo test --workspace — pass
  • compiled-engine Python parity — 340 passed
  • Rust core library — 1,045 passed, 1 ignored
  • comprehensive Python unit inventory — 5,001 passed; the three failures found were corrected and their committed-snapshot reruns pass
  • generator/sweeper/GDN focused suite — 24 passed
  • AIC sync renderer tests — 3 passed
  • CODEOWNERS policy — 99 passed; 2,750/2,750 tracked paths explicitly owned; generated outputs unchanged
  • Ruff check/format — pass (729 files)
  • release contract — built and verified exactly aisimulate-0.12.0-...whl plus aisimulate-core-0.12.0.crate

PRs migrated

#1578, #1551, #1591, #1557, #1478, #1441, #1506, #1590, #1598, #1554, #1581, #1542, #1574, #1559, #1576, #1482, #1558, #1562, #1471, #1533, #1594, #1600, and #1519.

Signed-off-by: Jason Zhou (Engrg-Hardware 1) <jasonzho@nvidia.com>
Signed-off-by: Jason Zhou (Engrg-Hardware 1) <jasonzho@nvidia.com>
@jasonqinzhou
jasonqinzhou requested review from a team as code owners September 1, 2026 18:21
@copy-pr-bot

copy-pr-bot Bot commented Sep 1, 2026

Copy link
Copy Markdown

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@jasonqinzhou
jasonqinzhou merged commit dd1e819 into main Sep 1, 2026
2 checks passed
@jasonqinzhou
jasonqinzhou deleted the jasonzho/sync-recent-aic-tail branch September 1, 2026 18:31
jasonqinzhou pushed a commit that referenced this pull request Sep 3, 2026
* feat(perfmodel): add GB200 multi-node communication data

Carry aiconfigurator#1593 into AISimulate with its measured GB200 multi-node NCCL and SGLang custom-allreduce rows plus TP16 golden coverage. AISimulate already has the generalized recorded-slice selection from #73, so the superseded Rust implementation is not duplicated.

The two parquet blobs are byte-identical to ai-dynamo/aiconfigurator commit aa4848d98d3ce14dba095a9a347ea7e877a6889c.

Signed-off-by: Yiming Liu <yimingl@nvidia.com>

* fix(perfmodel): select compatible communication fan-outs

Signed-off-by: Yiming Liu <yimingl@nvidia.com>

---------

Signed-off-by: Yiming Liu <yimingl@nvidia.com>
carlory pushed a commit to carlory/aisimulate that referenced this pull request Oct 30, 2026
* fix(migration): render path-rewritten renames

Signed-off-by: Jason Zhou (Engrg-Hardware 1) <jasonzho@nvidia.com>

* feat(migration): sync AIC changes through ce2824e

Signed-off-by: Jason Zhou (Engrg-Hardware 1) <jasonzho@nvidia.com>

---------

Signed-off-by: Jason Zhou (Engrg-Hardware 1) <jasonzho@nvidia.com>
carlory pushed a commit to carlory/aisimulate that referenced this pull request Oct 30, 2026
* feat(perfmodel): add GB200 multi-node communication data

Carry aiconfigurator#1593 into AISimulate with its measured GB200 multi-node NCCL and SGLang custom-allreduce rows plus TP16 golden coverage. AISimulate already has the generalized recorded-slice selection from ai-dynamo#73, so the superseded Rust implementation is not duplicated.

The two parquet blobs are byte-identical to ai-dynamo/aiconfigurator commit aa4848d98d3ce14dba095a9a347ea7e877a6889c.

Signed-off-by: Yiming Liu <yimingl@nvidia.com>

* fix(perfmodel): select compatible communication fan-outs

Signed-off-by: Yiming Liu <yimingl@nvidia.com>

---------

Signed-off-by: Yiming Liu <yimingl@nvidia.com>
Arsene12358 added a commit that referenced this pull request Oct 30, 2026
The layer-type fix moves 12 of the 24 odd Llama 4 layers from global to
chunked attention. The four Scout records still pass their 1% tolerance,
but Rule 1 of rust-core/parity.md asks a PR to refresh every record that
its change affects. All four surfaces moved between the live engine at
the base and at the branch head, so refresh all four with
pin_goldens.py --refresh:

- llama4-scout-b200-vllm-isl1024-osl2::static
- llama4-scout-b200-vllm-isl1024-osl2::mixed
- llama4-scout-b200-vllm-isl1024-osl2::agg
- llama4-scout-b200-vllm-isl1024-osl2::disagg

Largest moves, live at base to after: disagg_tpot 7.233 -> 7.217 ms
(-0.22%), static_gen 4.619 -> 4.615 ms (-0.09%), mixed_step
22.061 -> 22.059 ms (-0.01%). The diff touches 10 leaves and the 4
post_freeze_pins entries. The parity README gets a table with the
recorded, live-at-base and after values, and the Scout keys in its
refresh command.

The old records already differed from the live engine at the base, by up
to +0.19% in mixed_step. That gap predates this PR. The static and disagg
records were last pinned in PR #244. The mixed and agg records still hold
the values from the PR #73 sync. The gap stayed inside the 1% tolerance,
so no PR refreshed them. Their old pin commits are not in this
repository's history.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Signed-off-by: Yiming Liu <yimingl@nvidia.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants