Sync DeepSeek OCR HF job code

ds_batch_ocr/cli.py

@@ -8,7 +8,6 @@ from pathlib import Path
 from typing import Optional, Sequence
 
 from .config import ArtifactLocator, AssembleSettings, DescribeSettings, ExtractSettings, InferenceSettings
-from .logging_utils import configure_logging
 from .server import (
     DeepSeekClient,
     base_url_from_env,
@@ -115,7 +114,10 @@ def safe_max_tokens(desired: int, stage: str) -> int:
 
 
 def main(argv: Optional[Sequence[str]] = None) -> None:
-    configure_logging()
+    logging.basicConfig(
+        level=os.environ.get("LOG_LEVEL", "INFO").upper(),
+        format="%(asctime)s | %(levelname)s | %(name)s | %(message)s",
+    )
     args = parse_arguments(argv)
 
     stage = (args.stage or os.environ.get("PIPELINE_STAGE", "extract")).lower()
@@ -126,7 +128,7 @@ def main(argv: Optional[Sequence[str]] = None) -> None:
     base_url = base_url_from_env()
 
     launch_server = should_launch_server() and stage in {"extract", "describe"}
-    server_process: Optional[subprocess.Popen] = None
+    server_process: Optional["subprocess.Popen"] = None
 
     try:
         if launch_server:
@@ -357,8 +359,6 @@ def main(argv: Optional[Sequence[str]] = None) -> None:
             shutdown_server(server_process)
 
 
-import subprocess  # noqa: E402
-
 __all__ = ["main", "parse_arguments", "getenv_float", "getenv_int"]
 
 

ds_batch_ocr/config.py

@@ -104,7 +104,6 @@ class InferenceSettings:
 class ArtifactLocator:
     strategy: str = "local"
     repo_id: Optional[str] = None
-    repo_type: Optional[str] = None
     job_id: Optional[str] = None
     job_owner: Optional[str] = None
     uri: Optional[str] = None

ds_batch_ocr/hf_io.py

@@ -16,19 +16,6 @@ DEFAULT_CHUNK_MAX_FILES = 200
 DEFAULT_CHUNK_MAX_BYTES = 512 * 1024 * 1024
 
 
-def _read_positive_int_env(name: str, default: int) -> int:
-    raw = os.environ.get(name)
-    if not raw:
-        return default
-    try:
-        value = int(raw)
-        if value > 0:
-            return value
-    except ValueError:
-        pass
-    return default
-
-
 def _gather_files(output_dir: Path, path_in_repo: str) -> List[Tuple[Path, str, int]]:
     base = output_dir.resolve()
     entries: List[Tuple[Path, str, int]] = []
@@ -181,7 +168,6 @@ def maybe_upload_dataset(
     *,
     output_dir: Path,
     repo_id: Optional[str],
-    repo_type: str,
     path_in_repo: str,
     commit_message: Optional[str],
     revision: Optional[str],
@@ -197,8 +183,8 @@ def maybe_upload_dataset(
     token = env_or_none("HF_TOKEN")
     api = HfApi(token=token)
 
-    max_files = _read_positive_int_env("HF_UPLOAD_CHUNK_MAX_FILES", DEFAULT_CHUNK_MAX_FILES)
-    max_bytes = _read_positive_int_env("HF_UPLOAD_CHUNK_MAX_BYTES", DEFAULT_CHUNK_MAX_BYTES)
+    max_files = int(os.environ.get("HF_UPLOAD_CHUNK_MAX_FILES", DEFAULT_CHUNK_MAX_FILES))
+    max_bytes = int(os.environ.get("HF_UPLOAD_CHUNK_MAX_BYTES", DEFAULT_CHUNK_MAX_BYTES))
 
     files = _gather_files(output_dir, path_in_repo or "")
     if not files:
@@ -214,10 +200,10 @@ def maybe_upload_dataset(
         total_batches,
     )
 
-    LOGGER.info("Ensuring %s repo exists: repo_id=%s", repo_type, repo_id)
+    LOGGER.info("Ensuring dataset repo exists: repo_id=%s", repo_id)
     create_repo(
         repo_id=repo_id,
-        repo_type=repo_type,
+        repo_type="dataset",
         exist_ok=True,
         token=token,
     )
@@ -240,7 +226,7 @@ def maybe_upload_dataset(
         )
         api.create_commit(
             repo_id=repo_id,
-            repo_type=repo_type,
+            repo_type="dataset",
             revision=revision,
             operations=operations,
             commit_message=message,

ds_batch_ocr/stages.py

@@ -8,6 +8,7 @@ from typing import Any, Dict, List, Optional
 import shutil
 from datasets import load_dataset
 from PIL import Image, ImageOps
+from torch.utils.data import DataLoader
 
 from .config import (
     AssembleSettings,
@@ -28,6 +29,10 @@ from .hf_io import maybe_upload_dataset, resolve_stage_dir, env_or_none
 LOGGER = logging.getLogger(__name__)
 
 
+def _collate_single_item(batch: List[Any]) -> Any:
+    return batch[0]
+
+
 def run_stage_extract(settings: ExtractSettings) -> None:
     dataset = load_dataset(
         settings.dataset_name,
@@ -36,6 +41,20 @@ def run_stage_extract(settings: ExtractSettings) -> None:
         streaming=settings.stream_dataset,
     )
 
+    if settings.stream_dataset:
+        num_workers = max(0, getenv_int("EXTRACT_DATALOADER_WORKERS", 2))
+        prefetch_factor = max(1, getenv_int("EXTRACT_DATALOADER_PREFETCH", 2))
+        dataloader_kwargs: Dict[str, Any] = {
+            "batch_size": 1,
+            "num_workers": num_workers,
+            "collate_fn": _collate_single_item,
+        }
+        if num_workers > 0:
+            dataloader_kwargs["prefetch_factor"] = prefetch_factor
+        sample_iterator = iter(DataLoader(dataset, **dataloader_kwargs))
+    else:
+        sample_iterator = iter(dataset)
+
     settings.output_dir.mkdir(parents=True, exist_ok=True)
 
     documents: List[DocumentMetadata] = []
@@ -144,7 +163,7 @@ def run_stage_extract(settings: ExtractSettings) -> None:
         batch_contexts = []
         batch_requests = []
 
-    for idx, sample in enumerate(dataset):
+    for idx, sample in enumerate(sample_iterator):
         if settings.max_samples is not None and idx >= settings.max_samples:
             break
 
@@ -234,7 +253,6 @@ def run_stage_extract(settings: ExtractSettings) -> None:
     maybe_upload_dataset(
         output_dir=settings.output_dir,
         repo_id=settings.upload_repo_id,
-        repo_type=settings.upload_repo_type,
         path_in_repo=settings.upload_path_in_repo,
         commit_message=extract_commit,
         revision=settings.upload_revision,
@@ -443,7 +461,6 @@ def run_stage_describe(settings: DescribeSettings) -> None:
     maybe_upload_dataset(
         output_dir=settings.output_dir,
         repo_id=settings.upload_repo_id,
-        repo_type=settings.upload_repo_type,
         path_in_repo=settings.upload_path_in_repo,
         commit_message=describe_commit,
         revision=settings.upload_revision,
@@ -502,6 +519,27 @@ def run_stage_assemble(settings: AssembleSettings) -> None:
         markdown = stage1_doc_path.read_text(encoding="utf-8")
         enriched_markdown = enrich_markdown_with_captions(markdown, description_map)
 
+        document_with_boxes_relpath: Optional[str] = None
+        document_boxes_rel = document.get("document_with_boxes_path")
+        if document_boxes_rel:
+            source_boxes_path = stage1_dir / document_boxes_rel
+            if source_boxes_path.exists():
+                target_boxes_path = sample_output_dir / Path(document_boxes_rel).name
+                shutil.copy2(source_boxes_path, target_boxes_path)
+                document_with_boxes_relpath = (
+                    Path(sample_id) / target_boxes_path.name
+                ).as_posix()
+            else:
+                LOGGER.warning("Document image with boxes missing: %s", source_boxes_path)
+                failures.append(
+                    {
+                        "sample_id": sample_id,
+                        "dataset_index": document.get("dataset_index"),
+                        "missing_path": source_boxes_path.as_posix(),
+                        "reason": "document_with_boxes_missing",
+                    }
+                )
+
         final_doc_path = sample_output_dir / "document_final.md"
         write_text(final_doc_path, enriched_markdown)
 
@@ -551,6 +589,7 @@ def run_stage_assemble(settings: AssembleSettings) -> None:
                 "sample_id": sample_id,
                 "dataset_index": document.get("dataset_index"),
                 "final_document_path": final_doc_rel_path,
+                "document_with_boxes_image_path": document_with_boxes_relpath,
                 "figures": copied_figures,
             }
         )
@@ -561,6 +600,7 @@ def run_stage_assemble(settings: AssembleSettings) -> None:
                 "dataset_index": document.get("dataset_index"),
                 "document_markdown_path": final_doc_rel_path,
                 "document_markdown_text": enriched_markdown,
+                "document_with_boxes_image_path": document_with_boxes_relpath,
                 "figures": copied_figures,
             }
         )
@@ -581,7 +621,6 @@ def run_stage_assemble(settings: AssembleSettings) -> None:
     maybe_upload_dataset(
         output_dir=settings.output_dir,
         repo_id=settings.dataset_repo_id,
-        repo_type=settings.dataset_repo_type,
         path_in_repo=settings.dataset_path_in_repo,
         commit_message=assemble_commit,
         revision=settings.dataset_branch,
@@ -589,7 +628,6 @@ def run_stage_assemble(settings: AssembleSettings) -> None:
     publish_dataset_viewer_assets(
         dataset_records=dataset_records,
         repo_id=settings.dataset_repo_id,
-        repo_type=settings.dataset_repo_type,
         revision=settings.dataset_branch,
         commit_message=f"{assemble_commit} [dataset viewer]",
     )
@@ -648,18 +686,17 @@ def publish_dataset_viewer_assets(
     *,
     dataset_records: List[Dict[str, Any]],
     repo_id: Optional[str],
-    repo_type: str,
     revision: Optional[str],
     commit_message: str,
 ) -> None:
-    if not repo_id or repo_type.lower() != "dataset":
+    if not repo_id:
         return
     if not dataset_records:
         LOGGER.debug("No dataset records to publish for %s", repo_id)
         return
 
     try:
-        from datasets import Dataset, Features, Sequence, Value  # type: ignore
+        from datasets import Dataset, Features, Sequence, Value, Image  # type: ignore
     except Exception as exc:  # pragma: no cover - defensive logging
         LOGGER.warning("Datasets library unavailable; skipping viewer dataset publish: %s", exc)
         return
@@ -673,6 +710,7 @@ def publish_dataset_viewer_assets(
                 "dataset_index": int(record.get("dataset_index") or 0),
                 "document_markdown_path": str(record.get("document_markdown_path", "")),
                 "document_markdown_text": record.get("document_markdown_text", ""),
+                "document_with_boxes_image": record.get("document_with_boxes_image_path"),
                 "figures": [
                     {
                         "figure_id": str(fig.get("figure_id", "")),
@@ -685,6 +723,12 @@ def publish_dataset_viewer_assets(
         )
 
     dataset = Dataset.from_list(normalized)
+    if "document_with_boxes_image" in dataset.column_names:
+        try:
+            dataset = dataset.cast_column("document_with_boxes_image", Image())
+        except Exception as exc:  # pragma: no cover - defensive logging
+            LOGGER.warning("Failed to cast document_with_boxes_image column to Image: %s", exc)
+
     token = env_or_none("HF_TOKEN")
     try:
         dataset.push_to_hub(