polowitty/surfPRM-data
surfPRM-data SFT training data for surfPRM, a process reward model (PRM) for web agents. 9,821 pairwise step-preference samples stored as a single JSON array Each sample has one field conversation with system / user / assistant turns: the user turn contains the task intent, the current page's accessibility tree (AXTree), the previous action trajectory, the start/current URLs, and two candidate next actions; the assistant turn contains structured… See the full description on the dataset page: https://huggingface.co/datasets/polowitty/surfPRM-data.
027
surfPRM-data
SFT training data for surfPRM, a process reward model (PRM) for web agents.
- 9,821 pairwise step-preference samples stored as a single JSON array
- Each sample has one field
conversationwithsystem/user/assistantturns: the user turn contains the task intent, the current page's accessibility tree (AXTree), the previous action trajectory, the start/current URLs, and two candidate next actions; the assistant turn contains structured<State>/<Criteria>/<Analysis>/<Answer>reasoning ending in the preferred action - The exact prompt template is documented in the model cards of the models trained on this data: surfPRM-qwen3.5-4b and surfPRM-qwen3.5-9b
License
Released under Apache 2.0.