DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to fine-tune Laya on your own decisions for higher accuracy with Laya

Use the fine-tuning notebook or Apple Silicon script to train on decisions from your domain.

Short answer: Collect labeled decisions from your own workflow; then open the Fine-tuning notebook (Kaggle 2xT4) or the Apple Silicon script from the README; run the whole loop: build the dataset, train with RLCD, calibrate temperatures, evaluate and export. All steps.

Who it is forAn ML engineer whose zero-shot routing accuracy is not good enough for production.

The problemShipped checkpoints work zero-shot, but domain specific decisions need domain data to be reliable.

The outcomeA fine-tuned checkpoint trained with RLCD, calibrated and evaluated on your workflows.

What you need first

  • Python 3.10 or newer
  • pip or uv
  • A GPU is optional; checkpoints download on first use
  • Optional extras: laya[serve], laya[mcp], laya[langchain] and others

Steps

  1. Collect labeled decisions from your own workflow
  2. Open the Fine-tuning notebook (Kaggle 2xT4) or the Apple Silicon script from the README
  3. Run the whole loop: build the dataset, train with RLCD, calibrate temperatures, evaluate and export
  4. Compare accuracy against the base checkpoint on held-out decisions
  5. Load the exported checkpoint in your Router

Prompt for your agent

Help me fine-tune NandhaKishorM/laya on my labeled decisions in <file> using the README's Fine-tuning notebook flow. Build the dataset, train, calibrate and evaluate, then report accuracy against the base English checkpoint on a held-out split.

Caveat. The README's accuracy jump (0.362 to 0.766) is on its own typed-decisions benchmark; your gain depends on your data.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for NandhaKishorM/laya and "Fine-tune Laya on your own decisions for higher accuracy"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer