commit

Dockerfile

@@ -9,6 +9,8 @@ COPY ./app ./app
 COPY requirements.txt .
 
 RUN pip install --upgrade pip && pip install -r requirements.txt
+#Download spaCy English model
+RUN python -m spacy download en_core_web_sm
 
 # ✅ Fix: Use a safe cache directory
 ENV TRANSFORMERS_CACHE=/tmp/huggingface

app/main.py

@@ -1,4 +1,4 @@
-
+'''
 from fastapi import FastAPI
 from pydantic import BaseModel
 from app.model import model, tokenizer
@@ -48,3 +48,80 @@ If the user message or prompt is too long tell the user that you have received h
 @app.get("/")
 def read_root():
     return {"message": "Welcome to my Hugging Face Space!"}
+'''
+
+from fastapi import FastAPI
+from pydantic import BaseModel
+from app.model import model, tokenizer  # Ensure your model and tokenizer are imported from your app
+import torch
+import spacy
+
+app = FastAPI()
+
+# Load spaCy for name detection
+nlp = spacy.load("en_core_web_sm")
+
+class Prompt(BaseModel):
+    text: str
+
+# Function to extract user's name from message
+def extract_name(text: str) -> str:
+    doc = nlp(text)
+    for ent in doc.ents:
+        if ent.label_ == "PERSON":
+            return ent.text
+    lowered = text.lower()
+    if "my name is" in lowered:
+        return text.split("my name is")[-1].split(".")[0].strip().split()[0].capitalize()
+    elif "best," in lowered or "thanks," in lowered:
+        return text.strip().split()[-1].capitalize()
+    return ""
+
+@app.post("/generate")
+def generate(prompt: Prompt):
+    user_input = prompt.text.strip()
+    user_name = extract_name(user_input)
+
+    # Format the full prompt for the model
+    prompt_template = f"""
+You are a professional human email assistant working for a company.
+
+Your goal is to reply to user messages with helpful, professional, and clearly written email replies.
+
+Follow these rules:
+- NEVER mention you're an AI or a model.
+- Use complete, natural, and formal English — sound like a real assistant.
+- If the message includes a name (e.g., “My name is Grace” or ends with “Best, John”), politely address the person by that name.
+- Be brief and respectful if the request is unclear or general, e.g., "Could you please clarify your request?"
+- If the message contains specific details like claim numbers or appointment requests, acknowledge receipt and indicate further action will be taken.
+- If the message is too long or complex, thank the user and say you'll get back to them soon, addressing them by name if provided.
+- Always end with a polite closing, like "Best regards" or "Sincerely", without using placeholder names.
+
+Here is the user's message:
+\"\"\"{user_input}\"\"\"
+
+Reply with a professional email below. Do not include explanations, examples, or placeholders.
+"""
+    full_prompt = f"<s>[INST] {prompt_template.strip()} [/INST]"
+
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    inputs = tokenizer(full_prompt, return_tensors="pt").to(device)
+
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=256,
+            temperature=0.7,
+            top_p=0.9,
+            do_sample=True
+        )
+
+    # Decode and clean the result
+    response_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    clean_response = response_text.split("[/INST]")[-1].strip().strip('"').strip()
+
+    return {"response": clean_response}
+
+@app.get("/")
+def read_root():
+    return {"message": "Welcome to my Hugging Face Space!"}

requirements.txt

@@ -9,6 +9,6 @@ datasets==2.18.0
 fastapi==0.110.0
 uvicorn==0.25.0
 numpy<2.0
-
+spacy