Dataset Viewer
Auto-converted to Parquet Duplicate
The dataset viewer is not available for this split.
Parquet error: Scan size limit exceeded: attempted to read 723652725 bytes, limit is 300000000 bytes Make sure that 1. the Parquet files contain a page index to enable random access without loading entire row groups2. otherwise use smaller row-group sizes when serializing the Parquet files
Error code:   TooBigContentError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Terminal / Bash agent SFT mix

Merged corpus for post-training terminal agents (OpenEuroLLM). 47,986 rows. messages is a JSON string: list of {role, content}. Columns: id, source, license, kind, format, messages, task.

source rows kind license
Lite-Coder/LiteCoder-Terminal-SFT 11,255 trajectory MIT
prometheus04/microagent-train-v3 30,578 trajectory Apache-2.0
emirkaanozdemr/bash_command_data_6K 6,153 command_pair Apache-2.0

Trajectory formats differ (format column): Terminus-style JSON vs microagent <think>/<bash> XML. Filter or convert per format when training.

Caveats

  • Per-row license is preserved; check upstream cards for attribution.
  • Trajectories were generated by third-party LLMs (LiteCoder includes Claude Code and OpenHands runs; microagent synthetic rows use DeepSeek). Review those providers' terms for training use.
  • microagent-v3 synthetic rows are screened against Terminal-Bench 2.0; LiteCoder is trained/tuned toward Terminal-Bench, so evaluate with care for contamination.
  • Command pairs may include destructive commands; run only in sandboxes.
  • Excluded: microagent-v2 (NVIDIA license), nl2bash-custom (no license), shellsafe-bash-sft (gated), Nex agent-sft (~10 GB, mostly non-terminal), WebTerminal (pretraining text).
Downloads last month
97