Hkang/

adp-prm-trm-embeddings-qwen3-8b-domain500

Precomputed float32 pooling embeddings derived from neulab/agent-data-collection. The selection takes up to the first 500 standardized trajectories from each of 23 domains.

Model: Qwen/Qwen3-8B PRM: state/action pairs, 4,096-token head truncation TRM: full trajectories, 32,768-token tail truncation Hard negatives: none Reward labels: preserved when ADP details includes one; otherwise marked as assumed demonstration success

See… See the full description on the dataset page: https://huggingface.co/datasets/Hkang/adp-prm-trm-embeddings-qwen3-8b-domain500.