Update app.py

app.py

@@ -1,55 +1,15 @@
 import gradio as gr
-from transformers import pipeline, BitsAndBytesConfig, AutoModelForCausalLM, AutoTokenizer
 import PyPDF2
 import io
 import re
 import json
 import os
-import gc
-import torch
 from huggingface_hub import login
 from dotenv import load_dotenv
 
 # --- Configuration --- #
 load_dotenv()
-login(token=os.getenv("HF_TOKEN")) 
-
-# Quantization config (only used if CUDA is available)
-quant_config = BitsAndBytesConfig(
-    load_in_4bit=True,
-    bnb_4bit_compute_dtype=torch.float16,
-    bnb_4bit_quant_type="nf4"
-)
-
-# Check if CUDA is available
-cuda_available = torch.cuda.is_available()
-
-# Load tokenizer and model
-tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.3")
-if cuda_available:
-    # Use quantization if CUDA is available
-    model = AutoModelForCausalLM.from_pretrained(
-        "mistralai/Mistral-7B-Instruct-v0.3",
-        device_map="auto",
-        quantization_config=quant_config,
-        torch_dtype=torch.float16
-    )
-else:
-    # Fall back to full precision (no quantization) if no CUDA
-    model = AutoModelForCausalLM.from_pretrained(
-        "mistralai/Mistral-7B-Instruct-v0.3",
-        device_map="cpu",  # Explicitly set to CPU
-        torch_dtype=torch.float16
-    )
-
-# Initialize pipeline with preloaded model and tokenizer
-analyzer = pipeline(
-    "text-generation",
-    model=model,
-    tokenizer=tokenizer,
-    device_map="auto" if cuda_available else "cpu",  # Match model device
-    torch_dtype=torch.float16
-)
+login(token=os.getenv("HF_TOKEN"))
 
 # Skills set for faster lookups
 GENERAL_SKILLS = {
@@ -107,8 +67,8 @@ def calculate_scores(resume_text, job_desc=None):
 
     return scores, min(100, sum(scores.values()))
 
-def analyze_resume(pdf_file, job_desc=None):
-    """Optimized analysis with memory management"""
+def analyze_resume(pdf_file, job_desc=None, inference_fn=None):
+    """Analyze resume using Together AI inference"""
     resume_text = extract_text_from_pdf(pdf_file)
     scores, total_score = calculate_scores(resume_text, job_desc)
     
@@ -119,12 +79,8 @@ def analyze_resume(pdf_file, job_desc=None):
     Return ONLY valid JSON without markdown:"""
     
     try:
-        result = analyzer(
-            prompt,
-            max_new_tokens=300,
-            do_sample=False
-        )[0]["generated_text"]
-        
+        # Use Together AI inference function passed from gr.load
+        result = inference_fn(prompt)
         return {
             "score": {"total": total_score, "breakdown": scores},
             "analysis": json.loads(result),
@@ -134,11 +90,19 @@ def analyze_resume(pdf_file, job_desc=None):
         return {"error": str(e)}
 
 # --- Gradio Interface --- #
-with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.Markdown("""
-    <h1 style='font-size: 18px'>🎯 Resume Analyzer</h1>
-    <p><i>Optimized for memory efficiency</i></p>
-    """)
+with gr.Blocks(theme=gr.themes.Soft(), fill_height=True) as demo:
+    with gr.Sidebar():
+        gr.Markdown("# Resume Analyzer with Mistral-7B")
+        gr.Markdown("Powered by mistralai/Mistral-7B-Instruct-v0.3 via Together AI API. Sign in to use.")
+        button = gr.LoginButton("Sign in")
+    
+    # Load Mistral-7B from Together AI
+    inference = gr.load(
+        "models/mistralai/Mistral-7B-Instruct-v0.3",
+        accept_token=button,
+        provider="together",
+        _js="() => ({ max_new_tokens: 300, do_sample: false })"  # Pass generation params
+    )
     
     with gr.Row():
         with gr.Column(scale=1):
@@ -146,12 +110,11 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
                 gr.File(label="PDF Resume", type="binary"),
                 gr.Textbox(label="Job Description (Optional)", lines=3)
             ]
-            # Removed examples that required sample.pdf
         with gr.Column(scale=2):
             output = gr.JSON(label="Analysis")
     
     inputs[0].upload(
-        fn=analyze_resume,
+        fn=lambda pdf, job_desc: analyze_resume(pdf, job_desc, inference),
         inputs=inputs,
         outputs=output,
         queue=True