Upload folder using huggingface_hub

fined-tuned-model.lora.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6fc9f1de53fe3d4eee5c536a0d566dafaf1d11d0167c526506bc9d89c7c3ebe3
+size 5560280

flagged/log.csv

@@ -0,0 +1,4 @@
+Input,history,Response,history,flag,username,timestamp
+"am sad
+",,,,,,2024-09-04 18:19:14.583976
+,,"I'm sorry to you and your family. I'm sure this is very upsetting for you. I'm not sure I can really offer much help. I can only imagine how you feel. I wish I could offer you a hug. I'm glad you're considering counseling. That's a good sign. It's a good sign because you're here, reading this. It's a good sign you're here asking this question. It's a good sign you're looking for answers. It's a good thing you're here on this site. You're doing something good for yourself. You're taking care of you. You're looking for answers. You're asking questions. You're looking for help. You're looking for support. You're looking for friends. You're looking for someone to talk to. You're looking for someone to talk to. You're looking for someone to talk with. You're looking for someone to talk about it all. You're looking for someone to listen. You're looking for someone to talk to. You're looking for someone to talk with. You'",,,,2024-09-04 18:25:01.121662

main.py

@@ -3,58 +3,45 @@ import os
 import keras_nlp
 from transformers import AutoModelForCausalLM
 
-
 # Set Kaggle API credentials
-
 os.environ["KAGGLE_USERNAME"] = "rogerkorantenng"
 os.environ["KAGGLE_KEY"] = "9a33b6e88bcb6058b1281d777fa6808d"
 
 # Load LoRA weights if you have them
-LoRA_weights_path = "fined-tuned.lora.h5"
+LoRA_weights_path = "fined-tuned-model.lora.h5"
 gemma_lm = keras_nlp.models.GemmaCausalLM.from_preset("gemma_2b_en")
-
 gemma_lm.backbone.enable_lora(rank=4)  # Enable LoRA with rank 4
 gemma_lm.preprocessor.sequence_length = 512  # Limit sequence length
 gemma_lm.backbone.load_lora_weights(LoRA_weights_path)  # Load LoRA weights
 
 # Define the response generation function
-def generate_response(message, history):
+def generate_response(message):
     # Create a prompt template
     template = "Instruction:\n{instruction}\n\nResponse:\n{response}"
 
-    # Format the history and the current message into the prompt
-    formatted_history = ""
-    for user_msg, bot_msg in history:
-        formatted_history += template.format(instruction=user_msg, response=bot_msg)
-
-    # Add the latest message from the user
+    # Create the prompt with the current message
     prompt = template.format(instruction=message, response="")
-    print(prompt)
-
-    # Combine history with the latest prompt
-    final_prompt = formatted_history + prompt
-    print(final_prompt)
+    print("Prompt:\n", prompt)
 
     # Generate response from the model
-    response = gemma_lm.generate(final_prompt, max_length=256)
+    response = gemma_lm.generate(prompt, max_length=256)
     # Only keep the generated response
-    response = response.split("Response:")[1].strip()
+    response = response.split("Response:")[-1].strip()
 
-    print(response)
+    print("Generated Response:\n", response)
 
     # Extract and return the generated response text
     return response  # Adjust this if your model's output structure differs
 
 # Create the Gradio chat interface
-interface = gr.ChatInterface(
+interface = gr.Interface(
     fn=generate_response,  # Function that generates responses
-    chatbot=gr.Chatbot(height=300),  # Chatbot UI component
-    textbox=gr.Textbox(placeholder="Hello, am Sage, your mental health advisor", container=False, scale=7),
-    title="Local Model Chat Bot",
-    retry_btn=None,  # Disable retry button
-    undo_btn="Delete Previous",  # Enable undo button
-    clear_btn="Clear"  # Enable clear button
+    inputs=gr.Textbox(placeholder="Hello, I am Sage, your mental health advisor", lines=2, scale=7),
+    outputs=gr.Textbox(),
+    title="Welcome to Sage, your dedicated mental health advisor.",
+#     description="Chat with Sage, your mental health advisor.",
+#     live=True
 )
 
 # Launch the Gradio app
-interface.launch(share=True)
+interface.launch(share=True, share_server_address="hopegivers.tech:7000")

main.py.save

@@ -0,0 +1,52 @@
+import gradio as grimport os
+import keras
+import keras_nlp
+
+import os
+
+os.environ["KERAS_BACKEND"] = "jax"  
+# Avoid memory fragmentation on JAX backend.
+os.environ["XLA_PYTHON_CLIENT_MEM_FRACTION"]="1.00"
+
+import os
+
+# Set Kaggle API credentials
+os.environ["KAGGLE_USERNAME"] = "rogerkorantenng"
+os.environ["KAGGLE_KEY"] = "9a33b6e88bcb6058b1281d777fa6808d"
+
+# Load environment variables
+load_dotenv()
+
+# Replace this with the path or method to load your local model
+gemma_lm = keras_nlp.models.GemmaCausalLM.from_preset("gemma_2b_en")
+
+def generate_response(message, history):
+    # Format the conversation history for the local model
+    formatted_history = []
+    for user, assistant in history:
+        formatted_history.append(f"Instruction:\n{user}\n\nResponse:\n{assistant}")
+
+    # Add the latest user message to the history
+    formatted_history.append(f"Instruction:\n{message}\n\nResponse:\n")
+
+    # Join formatted history into a single string for input
+    input_text = "\n".join(formatted_history)
+
+    # Generate response from the local model
+    # Make sure to adjust this part according to your model's API
+    response = gemma_lm.generate(input_text, max_length=256)
+    
+    # Extract the response text
+    # Adjust the response extraction based on the actual structure of your model's output
+    return response[0]  # Change this line if necessary
+
+# Create the Gradio interface
+gr.ChatInterface(
+    generate_response,
+    chatbot=gr.Chatbot(height=300),
+    textbox=gr.Textbox(placeholder="You can ask me anything", container=False, scale=7),
+    title="Local Model Chat Bot",
+    retry_btn=None,
+    undo_btn="Delete Previous",
+    clear_btn="Clear"
+).launch(share=True)

requirements.txt

@@ -92,4 +92,5 @@ urllib3==2.2.2
 uvicorn==0.30.6
 websockets==12.0
 Werkzeug==3.0.4
-wrapt==1.16.0
+wrapt==1.16.0
+transformers