Update app.py

app.py

@@ -1,73 +1,83 @@
 # app.py - Fixed version with proper adapter loading
 import gradio as gr
 import torch
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 from peft import PeftModel, PeftConfig
 import os
 
 print("🚀 ATS Resume Optimizer - Starting...")
 
-# Load model
-print("\n📥 Loading tokenizer...")
-tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
-tokenizer.pad_token = tokenizer.eos_token
+# Check adapter files
+print("\n📋 Files in current directory:")
+for f in os.listdir("."):
+    print(f"  - {f}")
 
-print("📥 Loading base model (this takes 2-3 minutes)...")
-base_model = AutoModelForCausalLM.from_pretrained(
-    "mistralai/Mistral-7B-Instruct-v0.2",
-    torch_dtype=torch.float16,
-    device_map="auto",
-    low_cpu_mem_usage=True,
-)
+# Load model with proper config
+print("\n📥 Loading model configuration...")
 
-print("📥 Loading fine-tuned adapters...")
 try:
-    # Try loading with PeftConfig first
+    # Load PEFT config first to understand the adapter structure
     peft_config = PeftConfig.from_pretrained(".")
-    print(f"✅ Adapter config loaded: {peft_config.peft_type}")
+    print("✅ Adapter config loaded")
     
-    # Load the model with adapters
+    # Load tokenizer
+    print("\n📥 Loading tokenizer...")
+    tokenizer = AutoTokenizer.from_pretrained(peft_config.base_model_name_or_path)
+    tokenizer.pad_token = tokenizer.eos_token
+    print("✅ Tokenizer loaded")
+    
+    # Load base model
+    print("\n📥 Loading base model (this takes 2-3 minutes)...")
+    model = AutoModelForCausalLM.from_pretrained(
+        peft_config.base_model_name_or_path,
+        torch_dtype=torch.float16,
+        device_map="auto",
+        low_cpu_mem_usage=True,
+    )
+    print("✅ Base model loaded")
+    
+    # Load adapters with proper config
+    print("\n📥 Loading your fine-tuned adapters...")
     model = PeftModel.from_pretrained(
-        base_model,
+        model,
         ".",
-        is_trainable=False,
-        config=peft_config
+        config=peft_config,
     )
-    
-    # Merge adapters with base model for faster inference
-    print("🔄 Merging adapters with base model...")
-    model = model.merge_and_unload()
-    
     model.eval()
+    
     print("✅ Fine-tuned model loaded successfully!")
     MODEL_LOADED = True
     
 except Exception as e:
-    print(f"⚠️ Could not load fine-tuned adapters: {e}")
-    print("📝 This means the demo will use the base Mistral model")
-    print("   (not your fine-tuned version)")
-    model = base_model
-    model.eval()
+    print(f"❌ Error loading adapters: {e}")
+    print("\n⚠️ Falling back to base model only")
+    
+    # Fallback to base model
+    tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
+    tokenizer.pad_token = tokenizer.eos_token
+    
+    model = AutoModelForCausalLM.from_pretrained(
+        "mistralai/Mistral-7B-Instruct-v0.2",
+        torch_dtype=torch.float16,
+        device_map="auto",
+        low_cpu_mem_usage=True,
+    )
+    
     MODEL_LOADED = False
 
 def analyze_resume(resume_text, job_description):
     """Generate ATS analysis"""
     
     if not MODEL_LOADED:
-        return """⚠️ **Fine-Tuned Model Not Available**
+        return """⚠️ **Using Base Model Only**
         
-The adapter files are not loading correctly. The demo is currently using
-the base Mistral-7B model (not fine-tuned for ATS analysis).
+The fine-tuned adapters couldn't be loaded. The model will still work but responses may be less specific to ATS optimization.
 
-**For the developer:** 
-Check that adapter files are uploaded correctly:
-- adapter_config.json
-- adapter_model.safetensors (or .bin)
+To see the full fine-tuned version, please contact the developer.
 
-The adapters were saved from your training but aren't loading properly
-due to a path mismatch in the model structure.
+---
 
-**Temporary workaround:** Use the Colab Gradio demo instead, which works perfectly.
+**Analyzing with base Mistral-7B...**
 """
     
     if not resume_text or len(resume_text.strip()) < 50:
@@ -92,7 +102,10 @@ JOB DESCRIPTION:
     
     try:
         inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=2048)
-        inputs = {k: v.to(model.device) for k, v in inputs.items()}
+        
+        # Move to device
+        if torch.cuda.is_available():
+            inputs = {k: v.cuda() for k, v in inputs.items()}
         
         with torch.no_grad():
             outputs = model.generate(
@@ -102,17 +115,19 @@ JOB DESCRIPTION:
                 top_p=0.9,
                 do_sample=True,
                 pad_token_id=tokenizer.eos_token_id,
+                eos_token_id=tokenizer.eos_token_id,
             )
         
         response = tokenizer.decode(outputs[0], skip_special_tokens=True)
         
+        # Extract response
         if "[/INST]" in response:
             response = response.split("[/INST]")[1].strip()
         
         return response
     
     except Exception as e:
-        return f"❌ Error: {str(e)}\n\nTry with a shorter resume or job description."
+        return f"❌ Error: {str(e)}\n\nPlease try with shorter text."
 
 # Sample data
 SAMPLE_RESUME = """Sarah Johnson
@@ -122,17 +137,27 @@ PROFESSIONAL SUMMARY
 Software Engineer with 3+ years of experience in full-stack development.
 
 TECHNICAL SKILLS
-Python, JavaScript, TypeScript, React, Node.js, Git, Docker
+Languages: Python, JavaScript, TypeScript
+Frontend: React, HTML5, CSS3
+Backend: Node.js, Express
+Databases: PostgreSQL, MongoDB
+Tools: Git, Docker, AWS
 
-PROFESSIONAL EXPERIENCE
+EXPERIENCE
 
-Software Engineer | TechCorp Inc. | 2021 - Present
+Software Engineer | TechCorp | 2021 - Present
 • Built web applications serving 100K+ users
 • Improved performance by 40%
-• Worked in Agile teams
+• Implemented CI/CD pipelines
+• Collaborated in Agile teams
+
+Junior Developer | StartupXYZ | 2020 - 2021
+• Developed REST APIs
+• Created responsive UIs
+• Fixed bugs and added features
 
 EDUCATION
-BS in Computer Science | State University | 2020
+BS Computer Science | State University | 2020
 """
 
 SAMPLE_JOB = """Position: Senior Full Stack Developer
@@ -142,11 +167,16 @@ Required Skills:
 • Node.js, Express
 • MongoDB or PostgreSQL
 • REST API design
-• Git, Docker
-• AWS or cloud platforms
+• Git, Docker, AWS
 • Agile methodologies
 
 Experience: 3-5 years
+
+Responsibilities:
+• Design and develop web applications
+• Write clean, maintainable code
+• Code reviews and mentoring
+• Architecture decisions
 """
 
 # Gradio interface
@@ -155,31 +185,27 @@ with gr.Blocks(title="ATS Resume Optimizer", theme=gr.themes.Soft()) as demo:
     gr.Markdown("""
     # 🎯 ATS Resume Optimizer
     
-    ### Fine-Tuned Mistral-7B for Resume Analysis
+    ### AI-Powered Resume Analysis
     
+    Get instant feedback on your resume:
+    - ✅ **ATS Compatibility Score**
+    - 🔍 **Missing Keywords**
+    - 💡 **Optimization Suggestions**
     """)
     
     if not MODEL_LOADED:
         gr.Markdown("""
-        ⚠️ **Note:** The fine-tuned adapter is not loading properly. 
-        This demo is using the base model. For the full fine-tuned experience,
-        please use the [Colab demo](#).
+        > ⚠️ **Note:** Currently running with base model. Fine-tuned adapters couldn't be loaded.
+        > The tool will still provide useful analysis but may be less specific.
         """)
     
-    gr.Markdown("""
-    Get instant feedback on your resume:
-    - ✅ **ATS Compatibility Score**
-    - 🔍 **Missing Keywords**
-    - 💡 **Optimization Suggestions**
-    
-    ---
-    """)
+    gr.Markdown("---")
     
     with gr.Row():
         with gr.Column():
-            gr.Markdown("### 📄 Resume")
+            gr.Markdown("### 📄 Your Resume")
             resume_input = gr.Textbox(
-                label="Paste Your Resume",
+                label="Paste Resume",
                 placeholder="Copy and paste your resume...",
                 lines=12,
                 value=SAMPLE_RESUME
@@ -205,26 +231,32 @@ with gr.Blocks(title="ATS Resume Optimizer", theme=gr.themes.Soft()) as demo:
     
     gr.Markdown("""
     ---
-    ### 💡 About This Project
+    ### 💡 How to Use
+    
+    1. **Paste your resume** in the left box (or try the sample)
+    2. **Paste job description** in the right box
+    3. Click **"Analyze Resume"**
+    4. Wait 1-2 minutes for analysis
     
-    This tool was built by fine-tuning Mistral-7B using QLoRA on 2,000 resume-job pairs.
+    ### 🔬 About This Tool
     
-    **Technical Details:**
-    - Model: Mistral-7B-Instruct-v0.2 + QLoRA adapters
-    - Training: 250 steps, final loss 0.254
-    - Framework: Transformers, PEFT, Gradio
+    Built with Mistral-7B language model for intelligent resume analysis.
+    Identifies missing keywords and provides actionable suggestions.
     
-    **Note:** First analysis may take 2-3 minutes as the model loads into memory.
+    **First analysis takes longer** as the model loads into memory.
     
     ---
     
-    **Created by:** [Your Name] | [GitHub](#) | [LinkedIn](#)
+    💻 **Tech Stack:** PyTorch • Transformers • PEFT • Gradio  
+    🔗 **Links:** [GitHub](#) | [LinkedIn](#) | [Portfolio](#)
     """)
     
+    # Event
     analyze_btn.click(
         fn=analyze_resume,
         inputs=[resume_input, job_input],
         outputs=output
     )
 
+print("\n🚀 Launching Gradio interface...")
 demo.launch()