feat(dataset): make work share factor configurable with -f flag

nanochat/dataset.py

@@ -129,6 +129,15 @@ if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="Download FineWeb-Edu 100BT dataset shards")
     parser.add_argument("-n", "--num-files", type=int, default=-1, help="Number of shards to download (default: -1), -1 = disable")
     parser.add_argument("-w", "--num-workers", type=int, default=4, help="Number of parallel download workers (default: 4)")
+    parser.add_argument(
+        "-f",
+        "--work-share-factor",
+        type=int,
+        default=8,
+        help=(
+            """Controls how each worker's share of shards is subdivided. CHUNK_SIZE is computed as len(ids_to_download) // (num_workers * work_share_factor), so it is the number of tasks a worker pulls per request from the main process. for example, for 240 shards and 4 workers the default value (8) produces 7 shards per request. setting it 1 gives a worker its entire share (~60 shards) in one go with minimal coordination but slow progress updates. larger work-share-factor values make the main process hand out smaller batches more often for faster feedback at a small scheduling cost."""
+        ),
+    )
     args = parser.parse_args()
 
     num = MAX_SHARD + 1 if args.num_files == -1 else min(args.num_files, MAX_SHARD + 1)
@@ -137,7 +146,9 @@ if __name__ == "__main__":
     logger.info(f"Dataset target directory: {DATA_DIR}")
     logger.info(f"Dataset downloader debug logs will be written to: {log_path}")
 
-    CHUNK_SIZE = max(1, len(ids_to_download) // (args.num_workers * 8))
+    # pool.imap_unordered pulls `chunksize` tasks from the main process before asking for more
+    work_share_factor = max(1, args.work_share_factor)
+    CHUNK_SIZE = max(1, len(ids_to_download) // (args.num_workers * work_share_factor))
     ok_count = 0
     with Pool(processes=args.num_workers) as pool:
         for ok in tqdm(