Browser Agent ChatML Dialogues for SFT and RFT Training

Multi-turn dialogues formatted in ChatML style for browser agent tasks. The dataset is split into SFT and RFT training subsets. It was created by TIGER-Lab and published in October 2025.

Use Cases

• Training a browser agent to follow multi-step instructions using the 'messages' dialogue structure.
• Supervised fine-tuning (SFT) of language models on the 'sft.jsonl' split containing annotated conversation sequences.
• Reinforcement fine-tuning (RFT) using the 'rft.jsonl' split for reward modeling and policy optimization.
• Benchmarking agent performance on browser-related tasks by evaluating responses within the provided dialogue format.

Strengths

• Dataset is structured as JSON Lines for direct preview with the Hugging Face Hub Data Visualizer.
• Data is split into dedicated SFT and RFT subsets for specific training methodologies.
• Platform tags indicate the dataset size category is between 10K and 100K rows.

Limitations

• Specific row counts, column details, and sample data are unavailable for assessment.
• The dataset description is truncated, requiring users to visit the external page for full schema details.
• The license and exact file sizes are unknown, which may affect usage rights and logistics.

Provenance

Source

TIGER-Lab

Freshness

Last updated on October 31, 2025.