Fixed app v2

app.py

@@ -1,32 +1,26 @@
+from transformers import AutoTokenizer
+from optimum.intel.openvino import OVModelForCausalLM
 import gradio as gr
-from llama_cpp import Llama
 from transformers import pipeline
 import json
 
-# Load the Llama model
-try:
-    llm = Llama.from_pretrained(
-        repo_id="HuggingFaceTB/SmolLM2-360M-Instruct-GGUF",
-        filename="smollm2-360m-instruct-q8_0.gguf"  # Replace with the correct path to your GGUF file
-    )
-except Exception as e:
-    raise RuntimeError(f"Failed to load model: {e}")
+# Load OpenVINO GPT-J model
+model_id = "OpenVINO/gpt-j-6b-int4-ov"
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+model = OVModelForCausalLM.from_pretrained(model_id)
 
-# Load summarization model
+# Summarization pipeline
 summarizer = pipeline("summarization")
 
-# Summarize text to fit within token limits
 def summarize_text(text, max_length=100):
-    # Use the summarizer to condense the text
     summary = summarizer(text, max_length=max_length, min_length=25, do_sample=False)
     return summary[0]["summary_text"]
 
-# Function to match CV to job descriptions with debug information
 def match_cv_to_jobs(cv_text, job_descriptions):
     debug_info = "Debug Info:\n"
     results = []
     
-    # Summarize `cv_text` and `job_descriptions` to manage token limits
+    # Summarize the CV text
     summarized_cv = summarize_text(cv_text, max_length=400)
     debug_info += f"Summarized CV Text: {summarized_cv}\n"
     
@@ -35,7 +29,7 @@ def match_cv_to_jobs(cv_text, job_descriptions):
         summarized_description = summarize_text(description, max_length=100)
         debug_info += f"\nSummarized Job Description: {summarized_description}\n"
         
-        # Create a prompt to compare the summarized CV with each summarized job description
+        # Create a prompt to compare the summarized CV with each job description
         prompt = (
             f"Compare the following job description with this resume. Job Description: {summarized_description}. "
             f"Resume: {summarized_cv}. Provide a match score and a brief analysis."
@@ -43,21 +37,12 @@ def match_cv_to_jobs(cv_text, job_descriptions):
         debug_info += f"\nGenerated Prompt: {prompt}\n"
         
         # Generate response from the model
+        inputs = tokenizer(prompt, return_tensors="pt")
         try:
-            response = llm.create_chat_completion(
-                messages=[
-                    {
-                        "role": "user",
-                        "content": prompt
-                    }
-                ]
-            )
-            
-            # Extract the analysis text
-            response_content = response["choices"][0]["message"]["content"]
+            outputs = model.generate(**inputs, max_length=200)
+            response_content = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]
             debug_info += f"Model Response: {response_content}\n"
             
-            # Attempt to parse as JSON; if not JSON, use the raw text
             try:
                 response_data = json.loads(response_content)
                 results.append(response_data)

requirements.txt

@@ -5,3 +5,4 @@ torch==2.0.0  # Specifying PyTorch 2.0
 git+https://github.com/abetlen/llama-cpp-python.git
 nest_asyncio
 pydantic
+optimum