---
license: mit
datasets:
- K-and-K/knights-and-knaves
language:
- en
base_model:
- google/gemma-3-1b-it
pipeline_tag: text-generation
library_name: transformers
tags:
  - knights-and-knaves
  - gemma3
---


# Gemma-3-1B-IT — Knights-and-Knaves SFT

Paper Link: [https://arxiv.org/abs/2605.28814](https://arxiv.org/abs/2605.28814)

Cold-start supervised-fine-tuned (SFT) model of [`google/gemma-3-1b-it`](https://huggingface.co/google/gemma-3-1b-it) on the [Knights-and-Knaves](https://huggingface.co/datasets/K-and-K/knights-and-knaves) (K&K) logic-puzzle dataset. 

For the post-trained model on top of this SFT model, see [`Xkev/gemma-3-1b-it-kk-bes`](https://huggingface.co/Xkev/gemma-3-1b-it-kk-bes).

## Training

- **Base model**: `google/gemma-3-1b-it`
- **Dataset**: K&K puzzles (1k subset), formatted as chat with reasoning + JSON answer
- **Framework**: [verl](https://github.com/volcengine/verl) `sft_trainer`
- **Hyperparameters**: `lr=1e-5`, `weight_decay=0.01`, `lr_warmup_ratio=0.1`, cosine schedule, `epochs=3`, `dtype=bf16`

## Intended use

Research on logical reasoning and post-training. Not intended for general dialog or production.

## License

MIT. Base model `google/gemma-3-1b-it` is governed by Google's [Gemma Terms of Use](https://ai.google.dev/gemma/terms), which still apply to this model.