cyberneticlibrary

Fine-tune and adapt language models

fine-tuning-expertskillsetup L3★9,726

Jeffallan/claude-skills ↗

What it does

Adapt pretrained models to domain-specific tasks

Best for

When zero-shot performance is inadequate but full training is too expensive.

Inputs

· pretrained model
· domain dataset

Outputs

· fine-tuned weights
· validation metrics
· inference script

Requires

· Hugging Face
· PyTorch
· PEFT

Preconditions

· Base model selected
· labeled data

Failure modes

· Catastrophic forgetting
· overfitting to small dataset
· unstable loss

Trust signals

· Learning rate schedules
· validation curves
· parameter efficiency metrics