ci: bump pypa/gh-action-pypi-publish to v1.14.2 for metadata 2.5 - #135
Merged
Conversation
Hatchling now emits core packaging metadata v2.5, which the Twine bundled in v1.14.0 rejects (InvalidDistribution), breaking both the dev and release publish jobs. v1.14.2 bundles Twine v7, which accepts metadata 2.5. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
![macroscopeapp[bot]](https://avatars.githubusercontent.com/in/900172?s=60&v=4){kind=small}
ApprovabilityVerdict: Approved at Macroscope's review found this PR approvable — This is a small, self-contained CI dependency update in the existing development and release PyPI publishing jobs, enabling metadata 2.5 without changing triggers, permissions, artifacts, targets, or application behavior. Both changes are pinned to the same newer action revision. Notes:
You can add or adjust custom eligibility rules. Learn more. |
joanvelja
added a commit
to joanvelja/renderers
that referenced
this pull request
Aug 25, 2026
* feat: add Nemotron 3.5 (Lightning) renderer * chore: drop NVFP4 map entry, BF16 only * feat(laguna): route poolside/Laguna-S-2.1 to the XS-2.1 renderer S-2.1 is a larger sibling of XS-2.1 (48 layers, hidden 3072, top-10 of 256 experts) but ships a byte-identical chat_template.jinja and tokenizer (same special tokens/ids, same vocab). It reuses LagunaXS21Renderer via an exact-match MODEL_RENDERER_MAP entry; no new renderer/config/parser needed. Added to the shared render-parity matrices (config-parity, is_content, sampled_mask); token-exact parity against apply_chat_template verified for both enable_thinking polarities across all conversation shapes. * fix(laguna): give Laguna-S-2.1 its own renderer for its distinct template The initial routing (cdc158e) assumed S-2.1 shipped a byte-identical chat template to XS-2.1 and reused LagunaXS21Renderer. It doesn't: S-2.1 shares XS-2.1's tokenizer (same vocab/merges/special tokens) but its template defaults enable_thinking to True (XS-2.1 defaults False) and adds a preserve_thinking kwarg, widening the reasoning-display gate to `enable_thinking or preserve_thinking`. Reusing the XS-2.1 renderer mismatched apply_chat_template on the default flag (bare `</think>` vs empty `<think></think>` wrapper), failing test_render_ids / test_build_helpers. The config-parity barrage missed this because it always sets enable_thinking explicitly, never exercising the differing default. Add LagunaS21RendererConfig (enable_thinking=True, preserve_thinking=False) and LagunaS21Renderer, a thin LagunaXS21Renderer subclass that overrides only the reasoning-display gate. Route poolside/Laguna-S-2.1 to it. The token format is otherwise identical, so everything else is inherited. Verified token-exact against apply_chat_template for all enable_thinking × preserve_thinking combinations; added tests/test_laguna_s21.py for the S-2.1-specific behaviours the shared barrage can't reach. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs(laguna): list S-2.1 in the renderer config tables The renderer/config landed in the previous commit but was missing from the user-facing listings, which now cover its sibling M.1 (PrimeIntellect-ai#120): the README renderer roster, the per-renderer config table, and the thinking_retention default table. `LagunaS21RendererConfig.template_field_names()` is {enable_thinking, preserve_thinking}, and the renderer inherits XS.2's ``all`` retention default, so it joins that row. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * less yapping * refactor(laguna): move LagunaS21Renderer to its own module Review feedback: S-2.1 didn't belong in laguna_xs2.py. Pure code motion to renderers/laguna_s21.py, following the qwen35.py -> qwen36.py precedent for a cross-module renderer subclass. Updated the _LAZY_RENDERERS entry, the _populate_registry import, the config docstring cross-ref, and the test import. laguna_xs2.py still imports LagunaS21RendererConfig for two config-union annotations — the base __init__s must accept an S-2.1 config forwarded up through super(). Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * Add Gemma 4 renderer (PrimeIntellect-ai#123) * Add Gemma 4 renderer * Fix multimodal processor loading in CI * Fix Gemma 4 post-tool thinking parsing * fix(gemma4): classify media content parts before text `_emit_content` dispatched on `"text" in item` before `_is_image_part`. HF Arrow schema unification (`Dataset.from_list` over a heterogeneous content list) adds `text: None` to every image part, so any image that round-tripped through a dataset took the text branch, emitted nothing, and never expanded its soft tokens. Reorder to media-first, matching `qwen35.py` and `kimi_k25.py`. Two adjacent instances of the same hazard: - `_emit_tool_response_body` was safe from the zombie key (strict `== "text"`) but silently dropped *untyped* text parts; align it on the same dispatch order and text acceptance. - the system-content text-only guard accepted a schema-unified image part via `"text" in item` and emitted a stray space instead of raising; reject media parts explicitly. Normalizing each part to `dict` also lets `_is_image_part` (which requires `isinstance(item, dict)`) see Mapping-but-not-dict parts, which `_emit_content` otherwise accepts. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(gemma4): address renderer stability review * test(gemma4): carve out intentional template deviation * fix(gemma4): preserve post-tool token identity --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> * feat(renderers): add Inkling and Inkling-Small (PrimeIntellect-ai#103) * feat(renderers): add Inkling renderer (text + image + audio) Implement InklingRenderer for thinkingmachines/Inkling — a token-delimited vision+audio model whose chat template is a flat stream of role-tagged, <|end_message|>-terminated blocks (no BOS) with a one-off "Thinking effort level: {N}" system line and <|content_model_end_sampling|> as the assistant close/eos. - Text / tools / reasoning / reasoning_effort: byte-parity with apply_chat_template across all shapes (effort labels + raw floats, tool cycles, multi-tool-call, content=None, tool-name resolution via tool_call_id). - Image + audio: byte-parity with the native InklingProcessor, including the Qwen-VL-style placeholder expansion (image -> num_patches image pads, audio -> one audio pad per mel frame); ships pixel_values / audio_input_ids in MultiModalData with exact placeholder anchoring. - Full parse_response (native-JSON tool args, types preserved) and bridge_to_next_turn (mm carry-forward; refuses tool messages needing cross-turn name resolution). The InklingProcessor is native in transformers >= 5.14 (merged upstream 2026-07-15), so it loads without trust_remote_code. Bump the transformers floor to >=5.14.0 and fast-track it past the 7-day uv exclude-newer cooldown (drop the fast-track once 5.14.x clears the window). The full existing suite was re-validated on 5.14.1 with no regressions. Wiring: InklingRendererConfig (reasoning_effort str|float, validated), MODEL_RENDERER_MAP, MULTIMODAL_MODELS={"image","audio"}, registry, __init__ exports. Tests: tests/test_inkling.py plus Inkling rows in the shared render/config-parity/multimodal matrices (extended with an "inkling" family and an audio modality kit). Docs updated. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * style(renderers): trim narrative comments on Inkling wiring Condense the verbose rationale comments added with the Inkling renderer (MODEL_RENDERER_MAP / MULTIMODAL_MODELS entries, INKLING_EFFORT_MAP, the transformers pin, and the shared-test rows) to terse one-liners matching the surrounding style. Comments only — no behavior change. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(inkling): address Bugbot review — bridge mutation + audio hash - bridge_to_next_turn copied only the outer mm_data dicts, so appending a carried-forward modality's items mutated the caller's previous_multi_modal_data lists in place. Copy the per-modality lists too. Adds a regression test (test_inkling_bridge_does_not_mutate_prior_mm_data). - _audio_hash now folds in the sampling rate, so mm_hashes["audio"] uniquely identifies the clip (matching the processor cache key); identical samples at different rates no longer collide as media IDs. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(inkling): support Inkling-Small * fix(inkling): address compatibility review * fix(inkling): load input_image content parts --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Enforce explicit chat-template kwarg allowlists (PrimeIntellect-ai#126) * feat: add Qwen3.8 27B renderer (PrimeIntellect-ai#130) * ci: bump pypa/gh-action-pypi-publish to v1.14.2 for metadata 2.5 (PrimeIntellect-ai#135) --------- Co-authored-by: eexwhyzee <13672394+eexwhyzee@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: hallerite <git@hallerite.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
1 participant
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.
Why
The Publish Dev run on
main@ 1a5bd3b failed at the Publish to PyPI step:Hatchling (unpinned in
build-system.requires) now emits core packaging metadata v2.5, which the Twine bundled in gh-action-pypi-publish v1.14.0 rejects. The releasepublish.ymlbuilds the same way, so the nextrenderers-v*tag would fail identically.Fix
Bump the SHA pin in both
publish.ymlandpublish-dev.ymlto v1.14.2 (dc37677), which bundles Twine v7 with metadata 2.5 support (release notes).This unblocks the v0.1.10 release.
🤖 Generated with Claude Code
Note
Low Risk
CI-only dependency pin on the publish step; no application or packaging logic changes beyond restoring successful PyPI uploads.
Overview
Bumps the SHA-pinned
pypa/gh-action-pypi-publishaction from v1.14.0 to v1.14.2 in bothpublish-dev.ymlandpublish.yml.This fixes PyPI publish failures where builds produced by current Hatchling emit core metadata v2.5, which the Twine bundled in v1.14.0 rejects (
InvalidDistribution: '2.5' is not a valid metadata version). v1.14.2 ships Twine v7 with metadata 2.5 support, so dev publishes onmainand taggedrenderers-v*releases can upload again.Reviewed by Cursor Bugbot for commit d5b613c. Bugbot is set up for automated code reviews on this repo. Configure here.
Note
Bump
pypa/gh-action-pypi-publishto v1.14.2 in publish workflowsUpdates the pinned action SHA from v1.14.0 to v1.14.2 in both publish-dev.yml and publish.yml to support metadata 2.5. The publish step behavior is otherwise unchanged.
Macroscope summarized d5b613c.