inital commit

app.py

@@ -0,0 +1,157 @@
+import json
+import torch
+import time
+from docling.document_converter import DocumentConverter
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from pathlib import Path
+from fastapi import FastAPI, File, UploadFile, HTTPException
+import os
+from dotenv import load_dotenv
+import tempfile
+from supabase import create_client
+
+
+
+load_dotenv()
+app = FastAPI()
+
+model_name = "numind/NuExtract-1.5-tiny"
+device     = "mps" if torch.backends.mps.is_available() else "cpu"
+dtype      = torch.float16 if device=="mps" else torch.float32
+
+@app.on_event("startup")
+def startup_supabase():
+    global supabase
+    supabase = create_client(
+        os.getenv("DATABASE_URL"),
+        os.getenv("SUPABASE_SERVICE_ROLE_KEY")
+    )
+
+@app.on_event("startup")
+def load_model():
+    print("Loading model and tokenizer...", flush=True)
+    global model, tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+    model_name, torch_dtype=dtype, trust_remote_code=True
+    ).to(device).eval()
+    tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
+    print("Model and tokenizer loaded.", flush=True)
+
+def predict_NuExtract(texts, template, batch_size=10, max_length=5096, max_new_tokens=1024):
+    print("Starting NuExtract prediction...", flush=True)
+    start_time = time.perf_counter()
+    template_str = json.dumps(json.loads(template), indent=4)
+    prompts = [
+        "<|input|>\n"
+        "### Instruction:\n"
+        "Remplis la template JSON avec les informations extraits du texte.\n"
+        "Exemples types de formations :  CAP Boucherie, Licence Pro Métiers de l’Énergétique, Baccalauréat Général\n"
+        "Exemples catégories de formations :  Transport, énergie, langues, esthétique\n"
+        "Exemples mobilités :  permis B, permis C, permis D. si y'a juste la mention de permis on considère que c'est le permis B\n"
+        "Output *only* the completed JSON.\n"
+        "### Template:\n"
+        f"{template_str}\n"
+        "### Text:\n"
+        f"{text}\n\n"
+        "<|output|>"
+        for text in texts
+    ]
+    print("Prompts prepared.", flush=True)
+    outputs = []
+    with torch.no_grad():
+        for i in range(0, len(prompts), batch_size):
+            batch = prompts[i : i+batch_size]
+            enc = tokenizer(
+                batch,
+                return_tensors="pt",
+                truncation=True,
+                padding=True,
+                max_length=max_length
+            ).to(device)
+            print(f"Generating outputs with model for batch {i//batch_size+1}...", flush=True)
+            ids = model.generate(**enc, max_new_tokens=max_new_tokens, use_cache=False)
+            outputs += tokenizer.batch_decode(ids, skip_special_tokens=True)
+    print("Outputs generated.", flush=True)
+    elapsed = time.perf_counter() - start_time
+    print(f"NuExtract prediction completed in {elapsed:.2f} seconds.", flush=True)
+    return [out.split("<|output|>")[1] for out in outputs]
+
+template = """{
+    "nom": "", "email": "", "telephone": "",
+    "education": [{"type_de_formation": "", "categorie_de_formation": "", "annee_debut": "", "annee_fin": ""}],
+    "experience": [{"position": "", "entreprise": "", "annee_debut": "", "annee_fin": ""}],
+    "types_des_permis_de_conduire": [""]
+}"""
+
+data_model = {
+        "experience": [{"start_date": "", "end_date": "", "job_category_id": ""}],
+        "education": [{"training_type_id": "", "training_category_id": "", "start_date": "", "end_date": ""}],
+        "email": "",
+        "phone": "",
+        "mobility": [{"id": "", "title": ""}]
+}
+
+@app.post("/extract")
+async def extract(file: UploadFile = File(...)):
+    suffix = Path(file.filename).suffix or ".pdf"
+    try:
+        # Create one global client; reused across calls
+        # supabase = create_client(os.environ.get("DATABASE_URL"), os.environ.get("SUPABASE_SERVICE_ROLE_KEY"))
+
+        # Use Supabase client to query tables, map names to IDs
+        job_categories = supabase.table("Job_category").select("id, title").execute()
+
+        print("Job Categories: ", job_categories)
+
+        training_types = supabase.table("Training_type").select("id, title").execute()
+
+        print("Training Types: ", training_types)
+
+        training_categories = supabase.table("Training_category").select("id, title").execute()
+
+        print("Training Categories: ", training_categories)
+
+        mobility = supabase.table("Mobility").select("id, title").execute()
+
+        print("Mobility: ", mobility)
+
+        # … your LLM logic …
+            
+        with tempfile.NamedTemporaryFile(suffix=suffix, delete=False) as tmp:
+            data = await file.read()
+            tmp.write(data)
+            tmp_path = tmp.name
+        print(f"Upload saved to {tmp_path}", flush=True)
+    except Exception as e:
+        print(f"Cannot save upload: {e}", flush=True)
+        raise HTTPException(400, f"Cannot save upload: {e}")
+
+    try:
+        converter = DocumentConverter()
+        result    = converter.convert(tmp_path)
+        raw_text  = result.document.export_to_text()
+        print("Docling conversion complete.", flush=True)
+    except Exception as e:
+        print(f"Docling error: {e}", flush=True)
+        raise HTTPException(500, f"Docling error: {e}")
+    finally:
+        try: os.remove(tmp_path)
+        except OSError: pass
+
+    try:
+        extracted_json_str = predict_NuExtract([raw_text], template)[0]
+        print("Extraction with NuExtract complete.", flush=True)
+        print("⏺ RAW MODEL OUTPUT:\n", extracted_json_str)
+        print("⏺ RAW MODEL OUTPUT (repr):\n", repr(extracted_json_str))
+
+        print("Clearing Cache", flush=True)
+        if device == "mps":
+            torch.mps.empty_cache()
+        elif device == "cuda":
+            torch.cuda.empty_cache()
+        elif device == "cpu":
+            torch.cpu.empty_cache()
+        return {"result": json.loads(extracted_json_str)}
+    except Exception as e:
+        print(f"Extraction error: {e}", flush=True)
+        raise HTTPException(500, f"Extraction error: {e}")

requirements.txt

@@ -0,0 +1,86 @@
+annotated-types==0.7.0
+attrs==25.3.0
+beautifulsoup4==4.13.4
+certifi==2025.4.26
+charset-normalizer==3.4.2
+click==8.1.8
+dill==0.4.0
+docling==2.32.0
+docling-core==2.31.0
+docling-ibm-models==3.4.3
+docling-parse==4.0.1
+easyocr==1.7.2
+et_xmlfile==2.0.0
+filelock==3.18.0
+filetype==1.2.0
+fsspec==2025.3.2
+huggingface-hub==0.31.4
+idna==3.10
+imageio==2.37.0
+Jinja2==3.1.6
+jsonlines==3.1.0
+jsonref==1.1.0
+jsonschema==4.23.0
+jsonschema-specifications==2025.4.1
+latex2mathml==3.78.0
+lazy_loader==0.4
+lxml==5.4.0
+markdown-it-py==3.0.0
+marko==2.1.3
+MarkupSafe==3.0.2
+mdurl==0.1.2
+mpire==2.10.2
+mpmath==1.3.0
+multiprocess==0.70.18
+networkx==3.4.2
+ninja==1.11.1.4
+numpy==2.2.6
+opencv-python-headless==4.11.0.86
+openpyxl==3.1.5
+packaging==25.0
+pandas==2.2.3
+pillow==11.2.1
+pluggy==1.6.0
+pyclipper==1.3.0.post6
+pydantic==2.11.4
+pydantic-settings==2.9.1
+pydantic_core==2.33.2
+Pygments==2.19.1
+pylatexenc==2.10
+pypdfium2==4.30.1
+python-bidi==0.6.6
+python-dateutil==2.9.0.post0
+python-docx==1.1.2
+python-dotenv==1.1.0
+python-pptx==1.0.2
+pytz==2025.2
+PyYAML==6.0.2
+referencing==0.36.2
+regex==2024.11.6
+requests==2.32.3
+rich==14.0.0
+rpds-py==0.25.0
+rtree==1.4.0
+safetensors==0.5.3
+scikit-image==0.25.2
+scipy==1.15.3
+semchunk==2.2.2
+setuptools==80.8.0
+shapely==2.1.1
+shellingham==1.5.4
+six==1.17.0
+soupsieve==2.7
+sympy==1.14.0
+tabulate==0.9.0
+tifffile==2025.5.10
+tokenizers==0.21.1
+torch==2.7.0
+torchvision==0.22.0
+tqdm==4.67.1
+transformers==4.51.3
+typer==0.15.4
+typing-inspection==0.4.0
+typing_extensions==4.13.2
+tzdata==2025.2
+urllib3==2.4.0
+XlsxWriter==3.2.3