update space

app.py

@@ -1,4 +1,3 @@
-
 import os
 import tensorflow as tf
 from keras.models import load_model
@@ -9,8 +8,6 @@ from huggingface_hub import login
 from pymongo import MongoClient
 from transformers import AutoModelForCausalLM, AutoTokenizer
 from concurrent.futures import ThreadPoolExecutor
-import torch
-import random
 
 # Ensure the Hugging Face token is set
 tok = os.environ.get('HF_Token')
@@ -38,19 +35,16 @@ print("TensorFlow GPU Available:", tf.config.list_physical_devices('GPU'))
 from tensorflow.keras import mixed_precision
 
 if len(tf.config.list_physical_devices('GPU')) > 0:
-    try:
-        # Ensure the GPU supports mixed precision
-        gpu_device = tf.config.list_physical_devices('GPU')[0]
-        gpu_info = tf.config.experimental.get_device_details(gpu_device)
-        if 'compute_capability' in gpu_info and gpu_info['compute_capability'][0] >= 7:
-            # Set mixed precision policy to use float16 for better performance on supported GPUs
-            policy = mixed_precision.Policy('mixed_float16')
-            mixed_precision.set_global_policy(policy)
-            print("Using mixed precision with GPU")
-        else:
-            print("GPU does not support mixed precision or may not provide significant benefits. Using default precision.")
-    except Exception as e:
-        print(f"Error during mixed precision setup: {e}")
+    # Ensure the GPU supports mixed precision
+    gpu_device = tf.config.list_physical_devices('GPU')[0]
+    gpu_info = tf.config.experimental.get_device_details(gpu_device)
+    if 'compute_capability' in gpu_info and gpu_info['compute_capability'][0] >= 7:
+        # Set mixed precision policy to use float16 for better performance on supported GPUs
+        policy = mixed_precision.Policy('mixed_float16')
+        mixed_precision.set_global_policy(policy)
+        print("Using mixed precision with GPU")
+    else:
+        print("GPU does not support mixed precision or may not provide significant benefits. Using default precision.")
 else:
     print("Using CPU without mixed precision")
 
@@ -67,7 +61,8 @@ try:
         print(f"Models loaded successfully on {device_name}.")
 except Exception as e:
     print(f"Error loading models: {e}")
-    print("Falling back to CPU for model loading.")
+    if 'weight_decay' in str(e):
+        print("Invalid argument 'weight_decay' found. Please adjust optimizer settings.")
     my_model, auth_model = None, None
 
 # Updated Disease names and recommendations based on fecal analysis
@@ -93,7 +88,7 @@ class PoultryFarmBot:
             # Resize the image to match model input size (224x224)
             image_check = cv2.resize(image, (224, 224))
             # Add batch dimension to the image array
-            image_check = np.expand_dims(image_check, axis=0).astype('float16')
+            image_check = np.expand_dims(image_check, axis=0)
             print("Image preprocessing successful.")
             return image_check
         except Exception as e:
@@ -112,8 +107,7 @@ class PoultryFarmBot:
         # Predict using the fecal disease detection model
         try:
             print("Running model prediction...")
-            with tf.device('/GPU:0' if len(tf.config.list_physical_devices('GPU')) > 0 else '/CPU:0'):
-                indx = my_model.predict(image_check).argmax()
+            indx = my_model.predict(image_check).argmax()
             print(f"Prediction complete. Predicted index: {indx}")
             name = name_disease.get(indx, "Unknown disease")
             status = result.get(indx, "unknown condition")
@@ -161,32 +155,23 @@ bot = PoultryFarmBot()
 print("Loading Llama 3.2 model and tokenizer...")
 model_name = "meta-llama/Llama-3.2-1B"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
-try:
-    # Use a random seed to avoid potential CUDA random_device error
-    torch.manual_seed(random.randint(0, 10000))
-    model = AutoModelForCausalLM.from_pretrained(model_name).to('cuda' if torch.cuda.is_available() else 'cpu')
-    print("Llama 3.2 model and tokenizer loaded successfully.")
-except Exception as e:
-    print(f"Error loading Llama 3.2 model: {e}")
-    model = None
+model = AutoModelForCausalLM.from_pretrained(model_name)
+print("Llama 3.2 model and tokenizer loaded successfully.")
 
 # Set the padding token to EOS token or add a new padding token
 if tokenizer.pad_token is None:
     print("Adding pad token to tokenizer...")
     tokenizer.add_special_tokens({'pad_token': '[PAD]'})
-    if model:
-        model.resize_token_embeddings(len(tokenizer))
+    model.resize_token_embeddings(len(tokenizer))
     print("Pad token added and model resized.")
 
 
 # Define Llama 3.2 response generation
 def llama2_response(user_input):
-    if model is None:
-        return "Model is not available for generating response."
     try:
         print("Generating response using Llama 2...")
         # Tokenize user input for the Llama 2 model
-        inputs = tokenizer(user_input, return_tensors="pt", truncation=True, max_length=500, padding=True).to('cuda' if torch.cuda.is_available() else 'cpu')
+        inputs = tokenizer(user_input, return_tensors="pt", truncation=True, max_length=500, padding=True)
         # Generate a response using the Llama 2 model
         outputs = model.generate(
             inputs["input_ids"],
@@ -224,6 +209,7 @@ def chatbot_response(image, text):
         # Use Llama 3.2 for more accurate responses to user text queries
         return llama2_response(text)
 
+
 # Gradio interface styling and layout with ChatGPT-like theme
 print("Setting up Gradio interface...")
 with gr.Blocks(theme=gr.themes.Soft(primary_hue="green", neutral_hue="slate")) as chatbot_interface: