LLMModel Museum

Development Workflow

Core Data Flow

  1. cmd/generator fetches upstream model metadata from OpenRouter.
  2. The generator merges upstream data with local YAML overrides in models/.
  3. The merged registry is emitted into models_gen.go.
  4. cmd/translator incrementally fills missing description_cn values inside models/.
  5. After translation, rerun the generator so the new Chinese descriptions are embedded in models_gen.go.

Local Commands

The repository uses go-task as the main developer entry point.

task fmt
task lint
task test
task build
task generator
task translator
task releasecheck
task sync

Pass additional flags to the Go CLIs after --:

task generator -- -fetch-only
task generator -- -models-dir ./models -output-go ./models_gen.go
task translator -- -provider OpenAI -limit 50
task translator -- -dry-run -id-prefix qwen/

Generator Semantics

Translator Semantics

The scheduled GitHub Action no longer uses the retired GitHub Models inference endpoint. Configure any OpenAI-compatible provider with an encrypted repository secret and optional repository variables:

gh secret set LLM_API_KEY
gh variable set LLM_BASE_URL --body https://api.openai.com/v1
gh variable set LLM_MODEL --body gpt-4o-mini

When LLM_API_KEY is absent, AI model-card extraction and translation are skipped with an explicit workflow warning. The remaining deterministic sync pipeline continues to run.

Generated Files

Release Gating

Validation

Run this before sending a change:

task ci