Gemma 4 E2B Chessking (QLORA)
PEFT adapter trained on a bounded, game-level-split sample of short Lichess checkmates where the winner captured at most three opposing pieces. The sample contains 64 Luna-enriched games with a concrete real-world mapping and a video-scene description.
This repository contains an adapter, not a merged model or GGUF. Load it with the base model named above. The training data uses the checkpoint's own Gemma 4 chat template.
Held-out legal move rate: 1.000 (8 samples).
The exact source revision, selection rules, dataset hash, training settings,
hardware, package versions, and raw held-out replies are in run_manifest.json.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for ramonpzg/gemma-4-e2b-chessking
Base model
google/gemma-4-E2B Finetuned
google/gemma-4-E2B-it