slm-gemma-2b-raft-rlaif

Gemma 2 2B with RAFT (retrieval-augmented fine-tuning): gemma-2-2b-it -> QA SFT -> instruction SFT -> RLAIF v2 -> RAFT. ~14k examples of golden passage + 3 distractors (uniform golden position, 20% no-golden refusals).

Downloads last month
21
Safetensors
Model size
3B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support