| license: cc-by-nc-4.0 | |
| tags: [bagel, vlm-gym, world-model, pacman, dataset] | |
| # pacman_hard_cot_chunk_k5_train | |
| BAGEL VLM-Gym world-model dataset (pacman / cot). | |
| - CoT chunk-K train set: all-step interleaved imagined reasoning; re-grounds on the true frame every K=5 steps. | |
| - layout: Train-only. Gzipped-JSONL shards under `training/`; each row is one packed SFT sample with base64-JPEG frames inline. | |
| - images are base64-encoded JPEG frames stored inline in each JSONL row. | |
| Pairs with the matching `pacman` checkpoint(s) under the companion model org; CoT and non-CoT | |
| variants share the same underlying rollouts. | |