pass5 sha1 script 4c343f57

code/pass5/4c343f57/abeja_sha1.py

@@ -0,0 +1,51 @@
+"""abeja-cc-ja の各シャードの content SHA-1 を計算して reports/<prefix>/<dir>/<shard>.sha1.parquet に置く（完全一致重複の計画用）。"""
+from __future__ import annotations
+import argparse, hashlib, json, os, time
+from concurrent.futures import ProcessPoolExecutor
+from pathlib import Path
+
+
+def one(task: dict) -> dict:
+    import pyarrow as pa, pyarrow.parquet as pq
+    from huggingface_hub import hf_hub_download
+    t0 = time.perf_counter(); sh = task["shard"]
+    for k in range(6):
+        try: local = hf_hub_download(task["repo"], task["path"], repo_type="dataset", token=os.environ.get("HF_TOKEN"), local_dir=task["staging"]); break
+        except Exception as e:  # noqa: BLE001
+            if k == 5: return {"shard": sh, "error": repr(e)}
+            time.sleep(20 * (k + 1))
+    pf = pq.ParquetFile(local); shas = []
+    for b in pf.iter_batches(batch_size=4096, columns=["content"]):
+        shas.extend(hashlib.sha1((x or "").encode("utf-8")).digest() for x in b.column("content").to_pylist())
+    out = Path(task["out"]) / f"{sh}.sha1.parquet"; out.parent.mkdir(parents=True, exist_ok=True)
+    pq.write_table(pa.table({"sha1": pa.array(shas, pa.binary(20))}), out); Path(local).unlink(missing_ok=True)
+    return {"shard": sh, "rows": len(shas), "seconds": round(time.perf_counter() - t0, 1)}
+
+
+def main() -> None:
+    p = argparse.ArgumentParser(); p.add_argument("--repo", required=True); p.add_argument("--source-prefix", default="outputs_decontam"); p.add_argument("--dir", required=True)
+    p.add_argument("--out", type=Path, default=Path("/work/out")); p.add_argument("--staging", default="/work/staging"); p.add_argument("--limit-shards", type=int, default=None)
+    p.add_argument("--upload-repo", default=None); p.add_argument("--upload-prefix", default="reports/sha1_decontam"); p.add_argument("--workers", type=int, default=max(1, int(os.environ.get("CPU_CORES") or os.cpu_count() or 2) - 1))
+    a = p.parse_args()
+    from huggingface_hub import HfApi
+    api = HfApi(token=os.environ.get("HF_TOKEN"))
+    shards = sorted({e.path.split("/")[-1] for e in api.list_repo_tree(a.repo, path_in_repo=f"{a.source_prefix}/{a.dir}", repo_type="dataset", recursive=False)})
+    if a.limit_shards: shards = shards[: a.limit_shards]
+    tasks = [{"repo": a.repo, "path": f"{a.source_prefix}/{a.dir}/{sh}/{sh}.parquet", "shard": sh, "out": str(a.out / a.dir), "staging": a.staging} for sh in shards]
+    rows = 0; failed = []
+    with ProcessPoolExecutor(max_workers=a.workers) as ex:
+        for r in ex.map(one, tasks):
+            if "error" in r: failed.append(r["shard"]); print(json.dumps({"event": "error", **r}), flush=True); continue
+            rows += r["rows"]; print(json.dumps({"event": "done", **r}), flush=True)
+    (a.out / a.dir / "_summary.json").write_text(json.dumps({"dir": a.dir, "shards": len(tasks), "rows": rows, "failed": failed}), encoding="utf-8")
+    if a.upload_repo:
+        for k in range(6):
+            try: api.upload_folder(repo_id=a.upload_repo, repo_type="dataset", folder_path=str(a.out / a.dir), path_in_repo=f"{a.upload_prefix}/{a.dir}", commit_message=f"sha1 {a.dir}"); break
+            except Exception as e:  # noqa: BLE001
+                if k == 5: raise
+                time.sleep(60 * (k + 1))
+    print(json.dumps({"event": "finish", "dir": a.dir, "rows": rows, "failed": failed}), flush=True)
+
+
+if __name__ == "__main__":
+    main()