QVAC-18717 feat[mod]: add Gemma4-E2B/E4B/31B, Qwen3.5-0.8B/2B/4B/9B, Qwen3.6-27B/35B-A3B to registry#1988
Merged
yuranich merged 7 commits intoMay 12, 2026
Conversation
…-A3B models to registry
donriddo
changed the title
Replace /resolve/main/ with pinned commit hashes for all 36 new model entries — validate-json requires commit-pinned URLs.
donriddo
added
verify
safe-to-test
Gemma 4 (E2B, E4B, 31B) is released under Apache-2.0, not the custom Gemma Terms of Use used by earlier Gemma generations.
Reorder tags on all 36 new entries so the model family sits at tags[2] and "instruct" at tags[3], matching the established multimodal convention (e.g. Qwen3-VL). The previous order put "instruct" at tags[2], which the SDK naming pipeline treats as the family name and would generate colliding INST_* constants. Also correct params from "35B" to "35B-A3B" for the four Qwen3.6-35B-A3B MoE entries to preserve the variant qualifier.
yuranich
approved these changes
May 12, 2026
gianni-cor
approved these changes
May 12, 2026
Contributor
|
/review |
Contributor
Tier-based Approval Status |
Contributor
|
/review |
maxim-smotrov
approved these changes
May 12, 2026
Contributor
|
/review |
Proletter
pushed a commit
that referenced
this pull request
May 24, 2026
…Qwen3.6-27B/35B-A3B to registry (#1988) * feat[mod]: add Gemma4-E2B/E4B, Qwen3.5-0.8B/2B/4B/9B, Qwen3.6-27B/35B-A3B models to registry * feat[mod]: add Gemma4-31B-it models to registry * fix: pin HuggingFace source URLs to commit SHAs Replace /resolve/main/ with pinned commit hashes for all 36 new model entries — validate-json requires commit-pinned URLs. * fix[mod]: correct Gemma4 licenseId from gemma to Apache-2.0 Gemma 4 (E2B, E4B, 31B) is released under Apache-2.0, not the custom Gemma Terms of Use used by earlier Gemma generations. * fix[mod]: fix tag order and Qwen3.6-35B-A3B params Reorder tags on all 36 new entries so the model family sits at tags[2] and "instruct" at tags[3], matching the established multimodal convention (e.g. Qwen3-VL). The previous order put "instruct" at tags[2], which the SDK naming pipeline treats as the family name and would generate colliding INST_* constants. Also correct params from "35B" to "35B-A3B" for the four Qwen3.6-35B-A3B MoE entries to preserve the variant qualifier. --------- Co-authored-by: gianni-cor <gianfranco.cordella@tether.io>
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
4 participants
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.
What problem does this PR solve?
New multimodal models from Google (Gemma 4) and Qwen (3.5 and 3.6 series) are not yet in the registry.
How does it solve it?
Adds 36 entries to
packages/registry-server/data/models.prod.jsoncovering:All Gemma 4 entries use
licenseId: "Apache-2.0". Qwen entries uselicenseId: "Apache-2.0". All useengine: "@qvac/llm-llamacpp"with HuggingFace source URLs pinned to commit SHAs (fetched via HF API). All tagged["generation", "multimodal", "<family>", "instruct"]with the model family attags[2](gemma4/qwen3.5/qwen3.6), matching the existing multimodal convention.How was it tested?
validate-jsonCI check passes — 715 total entries, no syntax errors, all HF source URLs pinned to commit hashes.Breaking changes
None.