Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

Choose the model and effort per stage in lexvm generate

Open
  • Is something wrong?
#44 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
48/100
Issue type
Feature
Clarity
Clearly specified
Activity status
Active
Tech stack
typescript
Domain
backend, cli

Research direction

Start with packages/cli/src/generate/index.ts, llm.ts, and pipeline.ts to trace how model and effort options reach each stage; then read docs/authoring/pipeline.md. Add per-stage option and environment variable handling with global fallbacks, and check the fake LLM tests for stage selection and prompt caching. Done means the acceptance criteria pass and the docs describe recommended hybrid settings without hardcoded model IDs.

Written by the indexing model from the issue text.

Description

area: cli area: pipeline priority: medium type: feature

Context

The generation stages differ in what they demand. Concept and puzzle dependency chart need the strongest design judgment but few tokens; world, logic and repair stages produce most of the tokens and are checked by the verifier. Today lexvm generate uses one model for every stage. A hybrid (largest model for design, smallest capable model at high effort for writing and repair) should cut cost by an order of magnitude at similar quality, but needs per-stage configuration.

Scope

  • Options --model-design, --model-write, --model-repair (and matching --effort-*), falling back to --model / LEXVM_MODEL and --effort; equivalent environment variables.
  • No model ids hardcoded in the repository (existing rule); docs describe tiers, not ids.
  • design/run.json records per stage which option slot was used and token usage, without model ids (existing provenance rule).
  • Docs: recommended configurations by pack length, and the note that small models default to a lower effort and should run at high for writing and repair.

Acceptance criteria

  • Each stage uses the model and effort from its slot, verified with a fake LLM in tests.
  • Missing slot options fall back to the global options; missing global model still exits 2 with guidance.
  • Prompt caching still works within a stage group.
  • docs/authoring/pipeline.md documents the slots and a recommended hybrid configuration.

References

  • packages/cli/src/generate/index.ts
  • packages/cli/src/generate/llm.ts
  • packages/cli/src/generate/pipeline.ts
  • docs/authoring/pipeline.md
Dominant language
TypeScript
Stars
0
Forks
0
Avg merge
28m
Merged PRs (30d)
2

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from nullthrone/lexvm

All issues in nullthrone/lexvm

Similar issues

More TypeScript issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.