Skip to content

[CI][SM70] Sync #345 with latest main - #360

Merged
yangzhuxinyzx merged 51 commits into
codex/v100-qwen38-flash-next-nvfp4-20260826-140311from
agent/v100-fix-pr345-main-sync-20260827-164533
Aug 27, 2026
Merged

yangzhuxinyzx merged 51 commits into
codex/v100-qwen38-flash-next-nvfp4-20260826-140311from
agent/v100-fix-pr345-main-sync-20260827-164533

Conversation

@yangzhuxinyzx

Copy link
Copy Markdown
Contributor

Purpose

Synchronize #345 with the latest public main after #350 landed. The only textual conflict was in tests/benchmarks/test_benchmark_sm70_decode.py; both independent tests are retained.

Test Plan

Test Result

  • Full changed-file pre-commit: passed.
  • pytest -q --confcutdir=tests/benchmarks tests/benchmarks/test_benchmark_sm70_decode.py: 4 passed.
  • Final tree: exactly matches projected audit tree 0254c5343b3e29484ccb17a0b9e6fcbf21c9a675.

This PR contains no additional behavior change beyond carrying latest main into #345 and preserving both tests.

yangzhuxinyzx and others added 30 commits August 23, 2026 19:53
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Bumps [sentence-transformers](https://github.com/huggingface/sentence-transformers) from 5.2.0 to 6.0.0.
- [Release notes](https://github.com/huggingface/sentence-transformers/releases)
- [Commits](huggingface/sentence-transformers@v5.2.0...v6.0.0)

---
updated-dependencies:
- dependency-name: sentence-transformers
  dependency-version: 6.0.0
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <support@github.com>
Bumps [pillow](https://github.com/python-pillow/Pillow) from 10.4.0 to 12.3.0.
- [Release notes](https://github.com/python-pillow/Pillow/releases)
- [Changelog](https://github.com/python-pillow/Pillow/blob/main/CHANGES.rst)
- [Commits](python-pillow/Pillow@10.4.0...12.3.0)

---
updated-dependencies:
- dependency-name: pillow
  dependency-version: 12.3.0
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <support@github.com>
Bumps [black](https://github.com/psf/black) from 24.10.0 to 26.5.1.
- [Release notes](https://github.com/psf/black/releases)
- [Changelog](https://github.com/psf/black/blob/main/CHANGES.md)
- [Commits](psf/black@24.10.0...26.5.1)

---
updated-dependencies:
- dependency-name: black
  dependency-version: 26.5.1
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <support@github.com>
Freeze the public repository at the exact released v1.3.0 source tree before archiving public development. Ongoing work was mirrored to the personal private repository first; history is preserved without force-pushing.

Target-tag: v1.3.0
Target-tree: 4373fb8
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Bumps [webcolors](https://github.com/ubernostrum/webcolors) from 24.11.1 to 25.10.0.
- [Changelog](https://github.com/ubernostrum/webcolors/blob/trunk/docs/changelog.rst)
- [Commits](ubernostrum/webcolors@24.11.1...25.10.0)

---
updated-dependencies:
- dependency-name: webcolors
  dependency-version: 25.10.0
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <support@github.com>
Bumps [pandas](https://github.com/pandas-dev/pandas) from 2.2.3 to 2.3.3.
- [Release notes](https://github.com/pandas-dev/pandas/releases)
- [Commits](pandas-dev/pandas@v2.2.3...v2.3.3)

---
updated-dependencies:
- dependency-name: pandas
  dependency-version: 2.3.3
  dependency-type: direct:production
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>
Bumps [wrapt](https://github.com/GrahamDumpleton/wrapt) from 1.17.2 to 2.3.0.
- [Release notes](https://github.com/GrahamDumpleton/wrapt/releases)
- [Changelog](https://github.com/GrahamDumpleton/wrapt/blob/develop/docs/changes.rst)
- [Commits](GrahamDumpleton/wrapt@1.17.2...2.3.0)

---
updated-dependencies:
- dependency-name: wrapt
  dependency-version: 2.3.0
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <support@github.com>
Record the route-hit 73.645-token combined endpoint and passing quality gate, while identifying the FP8 route change that prevents assigning its delta to MXFP4. Continue follow-up work from the merged private main.

Co-authored-by: OpenAI Codex <codex@openai.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Keep the numerically sensitive shared-expert gate/up QPN8 route opt-in after its matched GSM8K endpoint regressed one item.

Co-authored-by: OpenAI Codex <codex@openai.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Keep the route opt-in until its matched full-model speed and quality gate closes.

Co-authored-by: OpenAI Codex <codex@openai.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Release per-expert conversion temporaries before pointer table construction and build strided pointer tables directly in torch-managed storage. This avoids unchecked cudaMallocAsync failures when V100 model loading exhausts driver headroom.

Co-authored-by: OpenAI Codex <codex@openai.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Remove the failed shared-gate prescaled integration, default PP2 TP4 QPN8 and packed FP13 off, and record the strict matched quality evidence. Preserve explicit opt-in compatibility for the existing diagnostic routes.

Co-authored-by: OpenAI Codex <codex@openai.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Correct the earlier same-accumulation assumption: the rejected shared-gate route selects a dedicated split-K-7 prescaled kernel whose reduction order does not match the control.

Co-authored-by: OpenAI Codex <codex@openai.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Decode ModelOpt E4M3FN scale bytes with integer and float32 Triton operations when native FP8 scale decoding is unavailable. Preserve signed zero in the manual branch so all finite scale encodings bit-match the reference across 2D/3D and FP16/BF16 outputs.

Assisted-by: OpenAI Codex
Signed-off-by: Leonccaa <166551845+Leonccaa@users.noreply.github.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
This reverts commit f4c3451.

Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
yangzhuxinyzx and others added 21 commits August 27, 2026 22:30
Restore the pre-freeze development tree and import the exact audited private main without rewriting history. The v1.3.0 tag remains unchanged.

Candidate and private source tree: bc484b5.

Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Signed-off-by: Leonccaa <166551845+Leonccaa@users.noreply.github.com>
…size

[Core][Performance] Avoid singleton hybrid KV cache group sizing
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
[Kernel][SM70] Extend unswizzled NVFP4 weight dequantization
…-20260827-153744

[Bugfix][SM70] Keep bounded FP13 path enabled
…llowup-20260826

[Bugfix][SM70] Gate PP2 TP4 fast paths on quality
…ers-6.0.0

Bump sentence-transformers from 5.2.0 to 6.0.0
…ne-20260823-115300

[Kernel][SM70] Pipeline batched E5M2 XQA decode
Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
…-20260827-161117

[CI][SM70] Fix #346 benchmark regex imports
…20260826-124851

[Bugfix][SM70] Audit DFlash2 output quality
…e-20260826-173751

[Bugfix][SM70] Restore DFlash2 prefill closure
…main-sync-20260827-164533

Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>

# Conflicts:
#	tests/benchmarks/test_benchmark_sm70_decode.py
@yangzhuxinyzx
yangzhuxinyzx marked this pull request as ready for review August 27, 2026 16:53
@yangzhuxinyzx
yangzhuxinyzx merged commit 4676f72 into codex/v100-qwen38-flash-next-nvfp4-20260826-140311 Aug 27, 2026
4 checks passed
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