Record 085 · Models · 2023
67%23.85.L
Alpaca
Stanford CRFM
“A Llama fine-tune that showed a weekend of self-instruct could make a helpful local model.”
Alpaca's 52K demonstrations and the ensuing replicas (Vicuna, GPT4All-LoRA) created the instruction-tuned local chat model.
Opened cheap instruction-tuning of local models.
- Self-instruct data
- 7B instruction model
- Replicable recipe
Instruction-following, from a 7B.
