Repository navigation
Promote SparkQwen 0.5.0 and SparkGLM 0.3.0 to the current recipes - #105
Merged
Merged
Conversation
Qualified on the Sparks on 2026-11-03: SparkQwen 0.5.0 at TP-2 and SparkGLM 0.3.0 at TP-2 and TP-4 give greedy output and prompt logprobs identical to the builds they replace, with speed level or better. - SparkQwen recipe v4 is current; v3 (0.4.0) is archived. - SparkGLM pair v3 (Enntity v80 drafter) is current; v1 and the never-promoted v2 are archived. v3 gets its own pin manifest, pins-0.3.0-v80.json, since its drafter differs from v2's. - SparkGLM TP-4 v2 is current; v1 (draft pins) is archived. Names, summaries, provenance and notes now describe what each installs. SparkQwen 0.5.0's packed cache was keyed by 0.4.0's image digest, which the generator had missed; it now uses its own. The recipe tests check each recipe against the manifest it names. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Qualified on the Sparks on 2026-11-03. Each build was compared on the same hardware with the build it replaces (receipts in sparkglm-research
experiments/qual-20261103):Promotion
linux-nvidia-dgx-spark-2x-qwen38-flash-next-atlas: v4 is current, v3 (0.4.0) archived.linux-nvidia-dgx-spark-2x-glm53-atlas: v3 (Enntity v80 drafter) is current, v1 and the never-promoted v2 archived. v3 getspins-0.3.0-v80.json, because its drafter differs from v2'spins-0.3.0.json.linux-nvidia-dgx-spark-4x-glm53-atlas-tp4: v2 is current, v1 (draft pins) archived.Names, summaries, provenance, hardware notes and both backend READMEs now describe what each recipe installs, not the old releases or "candidate".
Bug fixed on the way: SparkQwen 0.5.0's packed-weight cache path was keyed by 0.4.0's image digest, because the candidate generator replaced the image reference but not the bare digest. It now uses 0.5.0's.
Tests: the three recipe tests check each recipe against the manifest it names, and keep the default
pins.jsonas the verifier and installer fixture.recipe-versions.test.mjschecks that each promoted version is the default.check,lint,format:check,test:unit,package:checkandlloom recipe-indexpass.🤖 Generated with Claude Code