Modelpublic

anakin87/LFM2-2.6B-ttt-rl-merged

sourceHugging Faceotherupdated 6mo agoView on Hugging Face

0likes16downloads

Model Card

LFM2-2.6B-ttt-rl-merged

Merged standalone model after the first round of CISPO training for Tic Tac Toe.

This is the result of merging anakin87/LFM2-2.6B-ttt-rl (LoRA adapter) into anakin87/LFM2-2.6B-ttt-sft (SFT base). It serves as the base for the second round of RL training.

This is an intermediate checkpoint from 🎓 [LLM RL Environments Lil Course](https://github.com/anakin87/llm-rl-environments-lil-course), a hands-on course on building RL environments for Language Models, where models learn from rewards, not examples. It walks through the full process of turning a small open model into a specialist that outperforms a large proprietary one on a specific task (Tic Tac Toe). The final model is anakin87/LFM2-2.6B-mr-tictactoe.

🤗🕹️ [Play against the final model](https://huggingface.co/spaces/anakin87/LFM2-2.6B-mr-tictactoe)

Evaluation

100 games per setting.

Model vs random opponent	% Wins	% Draws	% Losses	% Follows format	% Games w invalid moves
LiquidAI/LFM2-2.6B	40	11	49	27.8	40
anakin87/LFM2-2.6B-ttt-sft	74	13	13	99.8	11
anakin87/LFM2-2.6B-ttt-rl-merged	86	12	2	100	1

Model vs optimal opponent	% Wins	% Draws	% Losses	% Follows format	% Games w invalid moves
LiquidAI/LFM2-2.6B	0	11	89	24.7	43
anakin87/LFM2-2.6B-ttt-sft	0	52	48	99	14
anakin87/LFM2-2.6B-ttt-rl-merged	0	85	15	100	1

Competent player, but still falls into fork traps against the optimal opponent.