Dom/Aivanov_scan_ogc
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(persist): corriger tag ocr_model et pipeline_version dans _meta

Browse Source 
Auparavant le JSON de sortie étiquetait systématiquement
`ocr_model: "zai-org/GLM-OCR"` et `pipeline_version: "v1"` alors que le
pipeline avait été basculé sur Qwen2.5-VL-3B en V2.

`_meta` lit désormais `MODEL_PATH` depuis `pipeline.ocr_qwen` pour
garantir la cohérence entre le modèle effectivement utilisé et la
trace dans le fichier.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Dom
2026-04-24 15:54:01 +02:00
parent 2ceb3c4916
commit 7dc3eba1fc
1 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
pipeline/persist.py
View File

@@ -9,9 +9,10 @@ DEFAULT_OUT = Path("output/v2")
def save_result(result: dict, out_dir: Path | str = DEFAULT_OUT) -> Path: def save_result(result: dict, out_dir: Path | str = DEFAULT_OUT) -> Path:
out_dir = Path(out_dir) out_dir = Path(out_dir)
out_dir.mkdir(parents=True, exist_ok=True) out_dir.mkdir(parents=True, exist_ok=True)
from .ocr_qwen import MODEL_PATH as OCR_MODEL_ID
result["_meta"] = { result["_meta"] = {
"pipeline_version": "v1", "pipeline_version": "v2",
"ocr_model": "zai-org/GLM-OCR", "ocr_model": OCR_MODEL_ID,
"generated_at": datetime.now(timezone.utc).isoformat(timespec="seconds"), "generated_at": datetime.now(timezone.utc).isoformat(timespec="seconds"),
} }
out_path = out_dir / f"{result['fichier']}.json" out_path = out_dir / f"{result['fichier']}.json"