pushed to hugging face

Dockerfile

@@ -0,0 +1,20 @@
+FROM python:3.10-slim
+
+RUN useradd -m -u 1000 user
+
+USER user
+
+ENV HOME=/home/user \
+    PATH=/home/user/.local/bin:$PATH
+
+WORKDIR $HOME/app
+
+COPY --chown=user requirements.txt .
+
+RUN pip install --no-cache-dir --upgrade -r requirements.txt
+
+COPY --chown=user . .
+
+EXPOSE 7860
+
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]

main.py

@@ -0,0 +1,93 @@
+from fastapi import FastAPI, HTTPException
+from fastapi.middleware.cors import CORSMiddleware
+from pydantic import BaseModel
+from transformers import AutoTokenizer, AutoModel
+from sklearn.cluster import KMeans
+import torch
+import numpy as np
+import spacy
+import time
+
+app = FastAPI(
+    title="Clinical Extractive Summarization",
+    description="SciBERT + KMeans NLP Engine for Medical Reports"
+)
+
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"], 
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+tokenizer = None
+model = None
+nlp = None
+
+class ReportRequest(BaseModel):
+    text: str
+    num_sentences: int = 3
+
+@app.post("/api/summarize")
+def summarize_medical_report(request: ReportRequest):
+    start_time = time.time()
+    
+    global tokenizer, model, nlp
+    if model is None:
+        print("Initializing SciBERT and SpaCy... This takes a moment.")
+        # Load SciBERT
+        model_name = "allenai/scibert_scivocab_uncased"
+        tokenizer = AutoTokenizer.from_pretrained(model_name)
+        model = AutoModel.from_pretrained(model_name)
+        try:
+            nlp = spacy.load("en_core_web_sm")
+        except OSError:
+            import spacy.cli
+            spacy.cli.download("en_core_web_sm")
+            nlp = spacy.load("en_core_web_sm")
+        print("Models loaded successfully!")
+
+    # 1. Safely split text into sentences using SpaCy NLP
+    doc = nlp(request.text)
+    sentences = [sent.text.strip() for sent in doc.sents if len(sent.text.strip()) > 5]
+    
+    # Edge case: Report is too short
+    if len(sentences) <= request.num_sentences:
+        return {"summary": request.text, "metadata": {"status": "too_short"}}
+
+    # 2. Get embeddings for each sentence using SciBERT
+    embeddings = []
+    for sent in sentences:
+        inputs = tokenizer(sent, return_tensors="pt", truncation=True, padding=True, max_length=512)
+        with torch.no_grad():
+            output = model(**inputs)
+        
+        # Extract the [CLS] token representation
+        cls_embedding = output.last_hidden_state[0][0].numpy()
+        embeddings.append(cls_embedding)
+
+    # 3. Use KMeans to cluster the embeddings and find the most central sentences
+    # n_init='auto' suppresses sklearn warnings
+    kmeans = KMeans(n_clusters=request.num_sentences, n_init='auto', random_state=42).fit(embeddings)
+    
+    avg = []
+    for i in range(request.num_sentences):
+        # Find the sentence closest to the cluster centroid
+        idx = np.argmin(np.linalg.norm(embeddings - kmeans.cluster_centers_[i], axis=1))
+        avg.append(idx)
+
+    # 4. Sort indices chronologically to maintain report flow
+    avg = sorted(list(set(avg)))
+    final_summary = " ".join([sentences[i] for i in avg])
+    
+    process_time = round((time.time() - start_time) * 1000, 2)
+    
+    return {
+        "summary": final_summary,
+        "metadata": {
+            "processing_time_ms": process_time,
+            "original_length": len(sentences),
+            "summary_length": len(avg),
+            "engine": "SciBERT + KMeans"
+        }
+    }

requirements.txt

@@ -0,0 +1,7 @@
+fastapi
+uvicorn
+pydantic
+transformers
+torch
+scikit-learn
+spacy