datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
opus-magnum-rl-eval
opus-magnum-rl-eval
Held-out evaluation logs for a 6-LoRA RL sweep on a hex-grid spatial planning benchmark inspired by Opus Magnum. Captures inspect_ai trajectories from base and LoRA-finetuned variants of three model families: Qwen 3.5 4B, Qwen 3.5 27B, and Kimi K2.6.
195 puzzles per evaluation × 12 model/representation combos = 2,340 trajectories. Each puzzle is held out from training (none of the (task_type, distance) cells in this set were used during RL).
Files… See the full description on the dataset page: https://huggingface.co/datasets/maxbittker/opus-magnum-rl-eval.hex-lora-opus-magnum-instructions-only-results
hex-lora-opus-magnum-instructions-only-results
Held-out evaluation logs for the same 6-LoRA RL sweep as
opus-magnum-rl-eval, but on a much harder eval task:
the 57-puzzle "instructions-only" set drawn from the
Opus Magnum campaign + curated
holdout puzzles. The agent runs an interactive Python REPL and must submit()
a working .solution file to the in-game verifier.
`57 puzzles × 6 epochs × (9 LoRA-sweep variants + 2 27B mt=4096 reruns
2 Gemini Flash baselines) = 4446 trajectories`.… See the full description on the dataset page: https://huggingface.co/datasets/robhaisfield/hex-lora-opus-magnum-instructions-only-results.moltbook-dataset
Moltbook: AI Agent Social Network Dataset
A large-scale dataset from Moltbook, a Reddit-style social platform designed for AI agents. The platform features community spaces called "submolts" (analogous to subreddits), where agents create posts, comment, upvote, and build karma. Human participation is not restricted.
This dataset captures a snapshot of the platform from its launch on January 27, 2026 through late March 2026.
Dataset Summary
Table
Records… See the full description on the dataset page: https://huggingface.co/datasets/opusmagnumown/moltbook-dataset.