app.py

import streamlit as st
from huggingface_hub import InferenceClient
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
import os
from PyPDF2 import PdfReader
import docx

def extract_cv_text(file):
    """Extract text from PDF or DOCX CV files."""
    if file is None:
        return "No CV uploaded"
    
    file_ext = os.path.splitext(file.name)[1].lower()
    
    if file_ext == '.pdf':
        reader = PdfReader(file)
        text = ""
        for page in reader.pages:
            text += page.extract_text()
        return text
    
    elif file_ext == '.docx':
        doc = docx.Document(file)
        text = ""
        for paragraph in doc.paragraphs:
            text += paragraph.text + "\n"
        return text
    
    else:
        return "Unsupported file format. Please upload PDF or DOCX files."

# Replace 'your_huggingface_token' with your actual Hugging Face access token
access_token = os.getenv('API_KEY')

# Initialize the tokenizer and model with the Hugging Face access token
tokenizer = AutoTokenizer.from_pretrained("google/gemma-2b-it", use_auth_token=access_token)
model = AutoModelForCausalLM.from_pretrained(
    "google/gemma-2b-it",
    torch_dtype=torch.bfloat16,
    use_auth_token=access_token
)
model.eval()  # Set the model to evaluation mode

# Initialize the inference client (if needed for other API-based tasks)
client = InferenceClient(token=access_token)

def conversation_predict(input_text):
    """Generate a response for single-turn input using the model."""
    # Tokenize the input text
    input_ids = tokenizer(f"""Job Description:
{input_text}

Instructions: Write a concise and professional email expressing interest in the position. 
Highlight relevant experience and skills from the CV that match the job requirements.
Keep the tone professional and enthusiastic.

Email:""", return_tensors="pt").input_ids

    # Generate a response with the model
    outputs = model.generate(input_ids, max_new_tokens=2048)

    # Decode and return the generated response
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

def respond(
    message: str,
    history: list[tuple[str, str]],
    system_message: str,
    cv_file,
    max_tokens: int,
    temperature: float,
    top_p: float,
):
    """Generate a response for a multi-turn chat conversation."""
    # Extract CV text and update system message
    cv_text = extract_cv_text(cv_file) if cv_file else "No CV provided"
    
    updated_system_message = f"""Task: Write a professional job application email.

CV Summary:
{cv_text}

{system_message}"""

    messages = [{"role": "system", "content": updated_system_message}]

    for user_input, assistant_reply in history:
        if user_input:
            messages.append({"role": "user", "content": user_input})
        if assistant_reply:
            messages.append({"role": "assistant", "content": assistant_reply})

    messages.append({"role": "user", "content": message})

    response = ""

    for message_chunk in client.chat_completion(
        messages=messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = message_chunk["choices"][0]["delta"].get("content", "")
        response += token
        yield response

# Streamlit UI
st.title("Job Application Email Generator")

# Instructions text area
system_message = st.text_area("System message", 
                              "Instructions: Write a concise and professional email expressing interest in the position.", 
                              height=150)

# CV file upload
cv_file = st.file_uploader("Upload CV (PDF or DOCX)", type=["pdf", "docx"])

# Sliders for max tokens, temperature, and top-p
max_tokens = st.slider("Max new tokens", min_value=1, max_value=2048, value=512, step=1)
temperature = st.slider("Temperature", min_value=0.1, max_value=4.0, value=0.7, step=0.1)
top_p = st.slider("Top-p (nucleus sampling)", min_value=0.1, max_value=1.0, value=0.95, step=0.05)

# Input message field
message = st.text_input("Job Description", "")

# Button to generate response
if st.button("Generate Email"):
    if message:
        response = conversation_predict(message)
        st.write("Generated Email:")
        st.write(response)
    else:
        st.warning("Please enter a job description.")