Skip to content

Add comprehensive fine-tuning plan for gpt-oss-20b wiki agent#8

Open
GhostScientist wants to merge 3 commits intomasterfrom
claude/finetune-gpt-wiki-agent-WVloy
Open

Add comprehensive fine-tuning plan for gpt-oss-20b wiki agent#8
GhostScientist wants to merge 3 commits intomasterfrom
claude/finetune-gpt-wiki-agent-WVloy

Conversation

@GhostScientist
Copy link
Copy Markdown
Owner

This adds detailed planning documents for fine-tuning OpenAI's gpt-oss-20b
to become a specialized architectural wiki agent for SemanticWiki local mode.

Documents include:

  • Dataset preparation: CodeWikiBench, DeepWiki, synthetic data generation
  • Fine-tuning execution: LoRA config, hyperparameters, training scripts
  • Evaluation: automated metrics, CodeWikiBench, task-specific evals

Target improvements over base model:

  • Source traceability: 50% → 90%+
  • Mermaid diagram validity: 70% → 95%+
  • Wiki completeness: 60% → 90%+

This adds detailed planning documents for fine-tuning OpenAI's gpt-oss-20b
to become a specialized architectural wiki agent for SemanticWiki local mode.

Documents include:
- Dataset preparation: CodeWikiBench, DeepWiki, synthetic data generation
- Fine-tuning execution: LoRA config, hyperparameters, training scripts
- Evaluation: automated metrics, CodeWikiBench, task-specific evals

Target improvements over base model:
- Source traceability: 50% → 90%+
- Mermaid diagram validity: 70% → 95%+
- Wiki completeness: 60% → 90%+
Adds QUICK-START.md with a minimal 3-4 hour plan:
- Uses only CodeWikiBench (no crawling/synthetic gen)
- Single epoch LoRA training
- Simple test script to verify output
- Cloud GPU options for those without local hardware
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.

2 participants