Skip to content

Dataset · saidutta69

SFT dataset from Claude Fable-5 traces

saidutta69/fable-5-premium

A cleaned English SFT dataset from Claude Fable-5 agent traces, 12,730 records.

What's inside

A high-quality English SFT dataset for text generation and token classification, compiled from Claude Fable-5 agent traces. The author emphasizes quality over quantity; the set underwent rigorous cleaning. Totaling 12,730 records, including train — 5,728, validation — 318, and test — 319; a newer fable-5-premium-v2 version exists.

What you can do with it

  • SFT a chat model on agent traces
  • Fine-tune LoRA for generation and token labeling
  • Verify SFT data cleaning quality
Tasks
Text generation, Token classification
Languages
English
Size
10K–100K rows
Downloads this month
1.1K
Likes
136

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("saidutta69/fable-5-premium")