Update handler.py

handler.py

@@ -2,7 +2,7 @@ import os
 from typing import Dict, List, Any
 from transformers import AutoTokenizer, AutoModelForCausalLM, AutoConfig
 import torch
-from unsloth import FastLanguageModel
+from peft import PeftModel  # Import PeftModel
 from dotenv import load_dotenv
 
 load_dotenv()
@@ -17,54 +17,55 @@ class EndpointHandler:
         max_new_tokens = int(os.getenv("MAX_NEW_TOKENS", 512))
         self.hf_token = os.getenv("HUGGINGFACE_TOKEN")
         self.model_dir = os.getenv("MODEL_DIR", ".")
+        self.base_model_name = "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B" # The base model ID
 
         print(f"MODEL_DIR: {self.model_dir}")
         print(f"Files in model directory: {os.listdir(self.model_dir)}")
 
         # --- 1. Load Config ---
-        # Load the configuration first, WITH trust_remote_code=True
-        self.config = AutoConfig.from_pretrained(self.model_dir, token=self.hf_token, trust_remote_code=True)
+        self.config = AutoConfig.from_pretrained(self.base_model_name, token=self.hf_token, trust_remote_code=True)
 
         # --- 2. Load Tokenizer ---
         try:
-            self.tokenizer = AutoTokenizer.from_pretrained(self.model_dir, token=self.hf_token, trust_remote_code=True)
+            self.tokenizer = AutoTokenizer.from_pretrained(self.base_model_name, token=self.hf_token, trust_remote_code=True)
         except Exception as e:
             print(f"Error loading tokenizer: {e}")
             raise
 
-        # --- 3. Load Model ---
+        # --- 3. Load Model with PeftModel ---
         try:
-            # Load base model, WITH trust_remote_code=True
-            self.model = AutoModelForCausalLM.from_pretrained(
-                self.config.base_model_name_or_path,
+            # Load base model
+             base_model = AutoModelForCausalLM.from_pretrained(
+                self.base_model_name,
                 config=self.config,
                 torch_dtype=torch.bfloat16,
                 token=self.hf_token,
                 device_map="auto",
-                trust_remote_code=True, #CRUCIAL
+                trust_remote_code=True,
             )
 
-            # Load and apply the LoRA adapter
-            self.model = FastLanguageModel.get_peft_model(self.model, self.model_dir)
-            FastLanguageModel.for_inference(self.model)
+            # Load the LoRA model using PeftModel
+            self.model = PeftModel.from_pretrained(base_model, self.model_dir)
+            # No need for FastLanguageModel.for_inference() here, PeftModel handles it
 
         except Exception as e:
             print(f"Error loading model: {e}")
             raise
 
-        # Define the prompt style
-        self.prompt_style = """Below is an instruction that describes a task, paired with an input that provides further context.
-Write a response that appropriately completes the request.
+                # Define the prompt style
+        self.prompt_style = """Below is an instruction that describes a task, paired with an input that provides further context. 
+Write a response that appropriately completes the request. 
 Before answering, think carefully about the question and create a step-by-step chain of thoughts to ensure a logical and accurate response.
 ### Instruction:
-You are BuildwellAI, an AI assistant specialized in UK building regulations and construction standards. You provide accurate, helpful information about building codes, construction best practices, and regulatory compliance in the UK.
-Always be professional and precise in your responses.
-
+You are BuildwellAI, an AI assistant specialized in UK building regulations and construction standards. You provide accurate, helpful information about building codes, construction best practices, and regulatory compliance in the UK. 
+Always be professional and precise in your responses. 
 ### Question:
 {}
 ### Response:
 <think>{}"""
 
+
+
     def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
         """
         Processes the input and generates a response.
@@ -82,7 +83,8 @@ Always be professional and precise in your responses.
         if torch.cuda.is_available():
             input_tokens = input_tokens.to("cuda")
 
-        with torch.no_grad():
+
+        with torch.no_grad():  # Ensure no gradient calculation
             output_tokens = self.model.generate(
                 input_ids=input_tokens.input_ids,
                 attention_mask=input_tokens.attention_mask,