Skip to content

fix(gallery): set MiniCPM5 context at the top level - #12163

Open
localai-org-maint-bot wants to merge 1 commit into
masterfrom
cron/model-gallery-20260920-run3
Open

localai-org-maint-bot wants to merge 1 commit into
masterfrom
cron/model-gallery-20260920-run3

Conversation

@localai-org-maint-bot

Copy link
Copy Markdown
Collaborator

Description

Fix the default context for the MiniCPM5-2B Q4_K_M and Q8_0 gallery entries. Both entries put context_size inside parameters, where schema.PredictionOptions silently ignores it. Move it to overrides.context_size so config.ModelConfig loads the intended 8,192 tokens, matching the F16 variant. Update the gallery documentation.

Notes for Reviewers

Validation:

  • Decoded gallery overrides into config.ModelConfig: Q4 and Q8 had nil ContextSize before the fix; all three builds now load 8192.
  • go test ./core/gallery -run TestGallery -ginkgo.focus='gallery/index.yaml|gallery variant lint helpers' -count=1 passes.
  • Full gallery suite: 392 passed, 3 failed, 9 skipped. The same three GitHub fixture failures reproduce on unchanged master.
  • All three existing SHA256 values match HF LFS metadata and x-linked-etag headers at publisher revision 2079a22f3beaa4e306449978533478fe0522f4b3. Weight URLs and hashes are unchanged.
  • Inference was not run.

This cron checked open gallery PRs and current HF trending models first. Existing green PRs were left untouched.

Signed commits

  • Human DCO sign-off required; no AI sign-off added.
  • Documentation updated (docs/content/) for user-facing changes.

The Q4 and Q8 overrides place context_size inside parameters, where
PredictionOptions ignores it. Move it beside parameters so both
builds use the intended 8,192-token context, matching F16.

Assisted-by: Codex:GPT-6
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