Skip to content

Support 24 more published ImageNet classifiers through YAML inheritance - #31

Merged
parkjinman98 merged 8 commits into
mainfrom
feat/published-classifier-coverage
Nov 4, 2026
Merged

parkjinman98 merged 8 commits into
mainfrom
feat/published-classifier-coverage

Conversation

@parkjinman98

@parkjinman98 parkjinman98 commented Nov 4, 2026 •

Copy link
Copy Markdown
Contributor

Summary

  • Add 24 already-published classifiers: EfficientViT (4), non-convolutional LeViT (5), and additional ViT RPN/RelPos/SRelPos/GAP/CLIP families (15).
  • Use a shared ordered models/bases/imagenet_classifier.yaml with only model-specific preprocessing/artifact overrides. Register task and top-level exports through the existing class factory.
  • Preserve exact mblt-model-ops recorded preprocessing and ImageNet dataset linkage; notably ViT_Large_Patch14_CLIP_224 uses TF, not OpenAI normalization.
  • All 24 repositories expose canonical MXQ but no ONNX. Add file_cfg.onnx_local_artifact_only: true on this new base: keep MXQ Hub loading, require trusted local ONNX, and skip only that framework in the metadata verifier. Positive/negative guard tests cover local ONNX, blocked remote ONNX, active MXQ, and unchanged marker-free downloads.
  • Existing 334 YAMLs and 595 variant resolved behavior remain unchanged. No CLI, OBB, version or HF repository/card changes.

Stack

Base: #29 (feat/hf-artifact-verification), itself on #28. The framework-specific verifier skip is introduced only here, after #29's metadata helper; this avoids editing the reviewed #29 head. No merges performed.

Source and revision policy

Recorded inputs and distribution contracts: mblt-model-ops commit 6fdbf471a3bd6ec22035a42c8730fe88a68881e4, models/image_classification/<Model>/_source/predict_float.input.json and distribution.yaml. New configs retain the established revision: main convention; historical approved artifact revisions are not recorded. The observed commits below are live audit diagnostics, not invented historical approval pins.

CPU validation

  • Full offline suite: 1478 passed, 19 deselected (11.48s).
  • Focused classifier, guard, metadata and legacy identity tests: 723 passed, 3 deselected.
  • Independent read-only ops preprocessing adapter execution for 24 models × 3 synthetic PNG shapes = 72 cases: max absolute error 2.384185791015625e-7, rtol=0, atol=5e-7. This compares preprocessing only, not inference weights or end-to-end accuracy. The bundle-local PIL import is shimmed; reference resize/crop/normalization operations run unchanged, parameterized by each exact distribution contract.
  • CPU test math, ordered YAML/recorded-input equality, registry discovery and top-level exports cover every addition.
  • Wheel builds; inspected all 358 model YAMLs plus the inherited base/ONNX marker; development assets absent. Ruff for new test/helper files and diff checks pass.
  • No GPU/NPU jobs, weight downloads or dataset downloads.

Actual HF file availability (all 24 queried)

Model MXQ under aries-rb Any ONNX published Observed commit
EfficientViT_B3 Yes No f04bbb0543b2becd767840558c7e9aa715121eb4
EfficientViT_L2 Yes No 81e7193f3b490051407ee6a3faa5143d86c6dd3a
LeViT_128S Yes No 21804ae0b9d93c4251e11e5c89bd683b14024583
EfficientViT_L1 Yes No c93cbc15d316aa581fedf12dd4edbc6e2d628c95
LeViT_128 Yes No 64ca04e2fd4cc81fabc1ae68d97e13d550b18460
LeViT_192 Yes No 5f48f8e459e572a8904c0a22d29f4c22dcd69b05
LeViT_256 Yes No e1a5f1af2b2db4ae2ab1f9882e26326401b56137
ViT_RelPos_Medium_Patch16_RPN_224 Yes No 41489de292c0d8a68b01d5f9dddb61c276e0eb34
EfficientViT_L3 Yes No 2fc5717c3023d069db3a120a032d6b40e3bdb837
ViT_RelPos_Base_Patch32_Plus_RPN_256 Yes No 7326cc3412ab1430c56782f638d50d87eb644bf4
ViT_Base_Patch16_RPN_224 Yes No 41b70f914862e372a42980a19134379098ae898b
ViT_Base_Patch16_CLIP_224 Yes No 7638ff9bbf3400ffc952ef4c3a2efae155b84e1a
LeViT_384 Yes No 872060d4caa777839b4fb2a4935caeb574fc9089
ViT_Base_Patch32_CLIP_448 Yes No 72a0af1bb9287552f6c7c4938f5a7e41e48ffb46
ViT_Medium_Patch16_GAP_256 Yes No 883b832554b3239dd87986f129c7150b5c822355
ViT_Large_Patch14_CLIP_224 Yes No ffd2d6accffa8445f71b1e7ebcf591b4c1705d3f
ViT_Base_Patch32_CLIP_224 Yes No 62cf9020cc06765f3a0601bdc394c2b4552eacd4
ViT_RelPos_Small_Patch16_224 Yes No 2877d86c5f8b6cb81d53460775d8561d0ba17955
ViT_Base_Patch32_CLIP_384 Yes No 0e6e9b5f2c1375058f1995e088d20280bc2af1a4
ViT_SRelPos_Medium_Patch16_224 Yes No 238b90395b01eb187b9727336bd131204b595110
ViT_Huge_Patch14_CLIP_224 Yes No 07555ac72b4880c1aa8ba6266dcc3f2e8616d64f
ViT_RelPos_Base_Patch16_224 Yes No 0e7fa05e8fee1a6272f356c70eb62a5a1fde8298
ViT_RelPos_Medium_Patch16_224 Yes No 1e5f404ce0cb49ef4e4b9e85f8b1644e10884955
ViT_SRelPos_Small_Patch16_224 Yes No a889bd57d5268dfb2b34204f3e93310687d424e3

No repository in this table contained any .onnx file under any path. ONNX requires a trusted local model_path/onnx_path; the API and derived filename convention remain intact.

Review guide

Head: b1dbd45926940fca4edec730ca07efe245d2bf8e. Check family inheritance order, the non-obvious TF normalization exception, all exports, board-specific canonical filenames, framework-local marker guarding only new models, and all-595 legacy identity coverage. Do not interpret observed HF tips as historical approval provenance.

🤖 Generated with Claude Code

Follow-up family-sibling commit

  • Add EfficientViT_B0/B1/B2 only, using official timm 1.0.14 r224 defaults. No model or weight download was needed.
  • artifact_status: pending_hf_family_sibling skips metadata network access and blocks automatic runtime/compile downloads; trusted local paths remain valid. Existing sibling YAML bytes stay unchanged.
  • LeViT is complete. Different ViT resolutions requiring squash-mode preprocessing and distinct EfficientViT M0–M5 architectures are deliberately excluded.
  • Final CPU suite: 1488 passed, 19 deselected; focused pending/compile/guard/all-595 identity 691 passed. Official timm preprocessing parity: 3 siblings × 3 RGB shapes, max absolute error 2.384185791015625e-7 (atol 5e-7). Wheel builds.
  • Security revision-policy thread remains open for maintainer decision; no claim that current-tip SHAs are approved historical provenance.

parkjinman98 and others added 5 commits November 4, 2026 16:22
Preserve existing alias and update semantics across 334 configs and 595 variants. Add ordered resolution and CPU preprocessing regression tests without changing shipped model definitions.

Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Verify exact runtime board paths using metadata only, preserve local-only restrictions and distinguish live commits from approval provenance.

Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Keep existing marker-free configs unchanged and preserve Hub MXQ loading. Skip only marked ONNX in metadata verification, with positive and negative CPU guard tests.

Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Nov 4, 2026 •

Copy link
Copy Markdown

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-11-04T09:40:12.730746Z b37c5bd Manual request
🔒 Security Review ✅ Completed 2026-11-04T09:41:26.580409Z b37c5bd Manual request

Security findings

Advisory findings (1)

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b1dbd45926

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread mblt_vision/wrapper.py Resolved
Exclude configuration-only marker from MXQ runtime kwargs, guard compilation downloads and cover local/unmarked positive paths.

Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
@parkjinman98

Copy link
Copy Markdown
Contributor Author

@codex review

Review guide for exact head 31ddef1a103793455476efbb0b082e00169b16be:

  • Fixed both first-round threads: configuration-only ONNX marker is excluded from MXQ backend kwargs, and direct compile resolution blocks remote ONNX for marked configs.
  • Positive/negative tests cover MXQ kwargs, explicit/configured trusted local ONNX, blocked remote ONNX, and unmarked download behavior.
  • Full offline CPU: 1480 passed, 19 deselected. Focused compile/guard/all-595 identity: 683 passed.
  • Both threads replied to and resolved. Please check complete runtime/compile propagation of this policy and unchanged marker-free behavior. No HF writes or new sibling configs in this fix; sibling pending-HF design is being confirmed separately.

@chatgpt-codex-connector

Copy link
Copy Markdown

Codex Review: Didn't find any major issues. Chef's kiss.

Reviewed commit: 31ddef1a10

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🛡️ Codex Security Review · Automatically triggered

Here are some automated security review suggestions for this pull request.

Reviewed commit: 31ddef1a10

ℹ️ About Codex security reviews in GitHub

This is an experimental Codex feature. Security reviews are triggered when:

  • You comment "@codex security review"
  • A regular code review gets triggered (for example, "@codex review" or when a PR is opened), and you’re opted in so security review runs alongside code review

Once complete, Codex will leave suggestions, or a comment if no findings are found.

Comment thread mblt_vision/models/EfficientViT_B3.yaml Resolved
Use official timm defaults and no-network pending status for new sibling configurations only. Preserve all pre-existing YAML bytes and allow trusted local artifacts without inventing HF publication metadata.

Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
@parkjinman98

Copy link
Copy Markdown
Contributor Author

@codex review

Exact head aaf0c14c2453e28c01be221744b56332f42dbe7d. Review guide: separate family-sibling commit adds only EfficientViT B0/B1/B2 with official timm preprocessing and pending-HF status, no automatic metadata/runtime/compile downloads, trusted-local positive tests, and backend kwargs filtering. Full offline CPU 1488 passed/19 deselected. Existing 334 YAMLs/595 variants stay identical. Prior runtime/compile findings are fixed/resolved. The security revision-policy finding has a factual provenance reply and remains open, pending maintainer decision; do not treat this as risk acceptance or approval-provenance pinning.

@chatgpt-codex-connector

Copy link
Copy Markdown

Codex Review: Didn't find any major issues. Breezy!

Reviewed commit: aaf0c14c24

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

@chatgpt-codex-connector

Copy link
Copy Markdown

🛡️ Codex Security Review · Automatically triggered

Security review completed. No security issues were found in this pull request.

Reviewed commit: aaf0c14c24

View security finding report

Only the user who started this review can view the report in Codex.

ℹ️ About Codex security reviews in GitHub

This is an experimental Codex feature. Security reviews are triggered when:

  • You comment "@codex security review"
  • A regular code review gets triggered (for example, "@codex review" or when a PR is opened), and you’re opted in so security review runs alongside code review

Once complete, Codex will leave suggestions, or a comment if no findings are found.

Retain pending and ONNX-only safety behavior plus new configured-board precedence from main.

Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
@parkjinman98

Copy link
Copy Markdown
Contributor Author

@codex review

Exact head b37c5bdf00f108da69967796c9eaf47ae9add4f1 after merging main 89a44a83 (squash #29), no force push. Main merge preserves classifier/pending/ONNX guards and brings configured-board precedence from merged #29. Resolved ancestry conflicts explicitly in guide, helper and tests; kept both sets of behavior. Maintainer decision (2026-11-04): main tracking is intentional; strict revision/digest pinning is not required. Security thread replied/resolved accordingly; no invented approval pins.

Full offline CPU 1489 passed,19 deselected. Includes all595 legacy variant identities. No version bump, release, tag or publish. Review exact head before squash merge.

@chatgpt-codex-connector

Copy link
Copy Markdown

Codex Review: Didn't find any major issues. Swish!

Reviewed commit: b37c5bdf00

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

@chatgpt-codex-connector

Copy link
Copy Markdown

🛡️ Codex Security Review · Automatically triggered

Security review completed. No security issues were found in this pull request.

Reviewed commit: b37c5bdf00

View security finding report

Only the user who started this review can view the report in Codex.

ℹ️ About Codex security reviews in GitHub

This is an experimental Codex feature. Security reviews are triggered when:

  • You comment "@codex security review"
  • A regular code review gets triggered (for example, "@codex review" or when a PR is opened), and you’re opted in so security review runs alongside code review

Once complete, Codex will leave suggestions, or a comment if no findings are found.

@parkjinman98

Copy link
Copy Markdown
Contributor Author

@codex review

One-time approval re-request for unchanged exact head b37c5bdf00f108da69967796c9eaf47ae9add4f1: code and security no-issues receipts are complete, but no qualifying approval reaction appeared within 30 minutes. No code changes since the tested head. Please confirm this exact head; merge remains held until explicit agent approval.

@chatgpt-codex-connector

Copy link
Copy Markdown

Codex Review: Didn't find any major issues. What shall we delve into next?

Reviewed commit: b37c5bdf00

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

@chatgpt-codex-connector

Copy link
Copy Markdown

🛡️ Codex Security Review · Automatically triggered

Security review completed. No security issues were found in this pull request.

Reviewed commit: b37c5bdf00

View security finding report

Only the user who started this review can view the report in Codex.

ℹ️ About Codex security reviews in GitHub

This is an experimental Codex feature. Security reviews are triggered when:

  • You comment "@codex security review"
  • A regular code review gets triggered (for example, "@codex review" or when a PR is opened), and you’re opted in so security review runs alongside code review

Once complete, Codex will leave suggestions, or a comment if no findings are found.

@parkjinman98
parkjinman98 merged commit f6673f6 into main Nov 4, 2026
3 checks passed
@parkjinman98
parkjinman98 deleted the feat/published-classifier-coverage branch November 4, 2026 09:53
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.

1 participant