wmh-bird-sql-traces / README.md
Silen's picture
update bird-sql dataset card
26e54b8 verified
|
Raw
History Blame Contribute Delete
1.8 kB
metadata
license: cc-by-sa-4.0
pretty_name: bird-sql agent-environment traces (world-model-harness)
language:
  - en
tags:
  - agent-trajectories
  - world-models
  - llm-environments

bird-sql — real agent-environment traces

Text-to-SQL over real SQLite databases: the agent explores a copy of the task's database and schema, then submits a SQL query.

Every trace is a REAL run: an LLM agent stepping against the actual benchmark environment, with each transition (tool call → true environment observation) recorded as OpenTelemetry GenAI spans (traces.otel.jsonl, one span per line). Captured by world-model-harness's environment-capture package, which also holds the adapter, capture scripts, and per-corpus provenance: see packages/environment-capture/bird-sql/.

License and attribution

Derived from bird-bench mini-dev (CC BY-SA 4.0); this corpus is redistributed under the same terms (cc-by-sa-4.0). The trace text embeds task data and environment output from the upstream benchmark — keep this attribution if you redistribute.

Contents

  • traces.otel.jsonl — the trace corpus (OTel GenAI spans, one JSON object per line)
  • data/ — task index (train/test splits: prompts + task metadata)
  • gold/ — per-task gold sidecars (graders read these; never staged into agent workspaces)
  • schemas/ — database DDL per task database

Using it

from huggingface_hub import hf_hub_download

path = hf_hub_download(
    "experiential-labs/wmh-bird-sql-traces", "traces.otel.jsonl", repo_type="dataset"
)

or, from a world-model-harness checkout:

uv run wmh download bird-sql