Upload export_runner.py with huggingface_hub

export_runner.py

@@ -0,0 +1,125 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.10"
+# dependencies = [
+#     "inspect-ai @ git+https://github.com/dvsrepo/inspect_ai.git@fallback-to-modified-for-hf-fs",
+#     "datasets",
+#     "huggingface_hub",
+#     "pandas",
+#     "pyarrow",
+# ]
+# ///
+"""Export-only runner that reads logs from a source dataset and exports to parquet."""
+
+import os
+import sys
+import tempfile
+from pathlib import Path
+
+from inspect_ai.analysis import evals_df, samples_df
+from huggingface_hub import HfApi
+
+
+def export_logs_to_parquet(log_dir: str, dataset_repo: str) -> None:
+    """Export eval logs to parquet format and upload to HuggingFace dataset.
+
+    Args:
+        log_dir: HF filesystem path to logs (e.g., "hf://datasets/username/name/logs")
+        dataset_repo: Dataset repository ID (e.g., "datasets/username/name")
+    """
+    # Get HF token from environment
+    hf_token = os.getenv("HF_TOKEN")
+    if not hf_token:
+        raise ValueError("HF_TOKEN environment variable not set")
+
+    api = HfApi(token=hf_token)
+
+    # Remove 'datasets/' prefix for API calls
+    repo_id = (
+        dataset_repo.replace("datasets/", "")
+        if dataset_repo.startswith("datasets/")
+        else dataset_repo
+    )
+
+    # Read evals dataframe
+    print("  Reading evals dataframe...")
+    print(f"    Log directory: {log_dir}")
+    try:
+        evals = evals_df(logs=log_dir)
+        print(f"    ✓ Read {len(evals)} eval records")
+    except Exception as e:
+        print(f"    ✗ Error reading evals: {e}")
+        raise
+
+    # Read samples dataframe
+    print("  Reading samples dataframe...")
+    try:
+        samples = samples_df(logs=log_dir)
+        print(f"    ✓ Read {len(samples)} sample records")
+    except Exception as e:
+        print(f"    ✗ Error reading samples: {e}")
+        raise
+
+    # Write to temporary parquet files
+    with tempfile.TemporaryDirectory() as tmpdir:
+        evals_path = Path(tmpdir) / "evals.parquet"
+        samples_path = Path(tmpdir) / "samples.parquet"
+
+        print(f"  Writing evals to parquet ({len(evals)} rows)...")
+        evals.to_parquet(evals_path, index=False, engine="pyarrow")
+
+        print(f"  Writing samples to parquet ({len(samples)} rows)...")
+        samples.to_parquet(samples_path, index=False, engine="pyarrow")
+
+        # Upload parquet files to root (HuggingFace will auto-detect as separate data files)
+        # We use descriptive names so they can be loaded separately
+        print("  Uploading evals.parquet...")
+        api.upload_file(
+            path_or_fileobj=str(evals_path),
+            path_in_repo="evals.parquet",
+            repo_id=repo_id,
+            repo_type="dataset",
+            token=hf_token,
+        )
+
+        print("  Uploading samples.parquet...")
+        api.upload_file(
+            path_or_fileobj=str(samples_path),
+            path_in_repo="samples.parquet",
+            repo_id=repo_id,
+            repo_type="dataset",
+            token=hf_token,
+        )
+
+    print(
+        f"  ✓ Parquet files available at: https://huggingface.co/datasets/{repo_id}/tree/main"
+    )
+
+
+if __name__ == "__main__":
+    if len(sys.argv) < 3:
+        print("Usage: export_runner.py <source_dataset_repo> <target_dataset_repo>")
+        print("  source_dataset_repo: Dataset repo to read logs from (e.g., datasets/username/name)")
+        print("  target_dataset_repo: Dataset repo to export parquet to (e.g., datasets/username/name)")
+        sys.exit(1)
+
+    source_dataset_repo = sys.argv[1]
+    target_dataset_repo = sys.argv[2]
+
+    # Construct log directory path for HF filesystem
+    if not source_dataset_repo.startswith("datasets/"):
+        source_dataset_repo = f"datasets/{source_dataset_repo}"
+    log_dir = f"hf://{source_dataset_repo}/logs"
+
+    print(f"Exporting logs from: {log_dir}")
+    print(f"Target dataset: {target_dataset_repo}")
+
+    try:
+        export_logs_to_parquet(log_dir, target_dataset_repo)
+        print("\n✓ Export completed successfully!")
+    except Exception as e:
+        import traceback
+        print(f"\n✗ Export failed: {e}")
+        print(f"\nFull traceback:")
+        print(traceback.format_exc())
+        sys.exit(1)