Update app.py

app.py

@@ -5,44 +5,61 @@ Diabetes Version
 @email: [email protected]
 """
 
-# Import necessary libraries
 import streamlit as st
-from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, pipeline
 from openai import OpenAI
 import os
-import torch
-from peft import LoraConfig, PeftModel, prepare_model_for_kbit_training, get_peft_model
-from huggingface_hub import login
+import sys
+from dotenv import load_dotenv, dotenv_values
+load_dotenv()
 
-# Initialize the OpenAI client
+
+# initialize the client
 client = OpenAI(
     base_url="https://api-inference.huggingface.co/v1",
-    api_key=os.environ.get("HUGGINGFACEHUB_API_TOKEN"),
+    # "hf_xxx" # Replace with your token
+    api_key=os.environ.get('HUGGINGFACEHUB_API_TOKEN')
 )
 
-api_token = os.getenv("HUGGINGFACEHUB_API_TOKEN")
-if api_token:
-    login(token=api_token)
-else:
-    st.error("API token is not set in the environment variables.")
-
-# Define model links
+# Create supported models
 model_links = {
-    "HAH-2024-v0.1": "drmasad/HAH-2024-v0.11"
+    "HAH v0.1": "drmasad/HAH-2024-v0.11",
+    "Mistral": "mistralai/Mistral-7B-Instruct-v0.2",
 }
 
-# Set selected model
-selected_model = "HAH-2024-v0.1"
+# Pull info about the model to display
+model_info = {
+    "HAH v0.1":
+        {'description': """HAH 0.1 is a fine tuned model based on Mistral 7b instruct.\n \
+            \nIt was created by Dr M. As'ad using 250k dB rows sourced from open source articles on diabetes** \n""",
+         'logo': 'https://www.hmgaihub.com/untitled.png'},
+    "Mistral":
+        {'description': """The Mistral model is a **Large Language Model (LLM)** that's able to have question and answer interactions.\n \
+            \nIt was created by the [**Mistral AI**](https://mistral.ai/news/announcing-mistral-7b/) team as has over  **7 billion parameters.** \n""",
+         'logo': 'https://mistral.ai/images/logo_hubc88c4ece131b91c7cb753f40e9e1cc5_2589_256x0_resize_q97_h2_lanczos_3.webp'},
+
+}
 
-# Display welcome message
-st.title("Welcome to HAH-2024-v0.1")
 
-# Sidebar setup
-temp_values = st.sidebar.slider("Select a temperature value", 0.0, 1.0, (0.5))
 def reset_conversation():
+    '''
+    Resets Conversation
+    '''
     st.session_state.conversation = []
     st.session_state.messages = []
+    return None
+
+
+# Define the available models
+models = [key for key in model_links.keys()]
+
+# Create the sidebar with the dropdown for model selection
+selected_model = st.sidebar.selectbox("Select Model", models)
 
+# Create a temperature slider
+temp_values = st.sidebar.slider('Select a temperature value', 0.0, 1.0, (0.5))
+
+
+# Create model description
 st.sidebar.button("Reset Chat", on_click=reset_conversation)
 st.sidebar.write(f"You're now chatting with **{selected_model}**")
 st.sidebar.image("https://www.hmgaihub.com/untitled.png")
@@ -50,89 +67,61 @@ st.sidebar.markdown("*Generated content may be inaccurate or false.*")
 st.sidebar.markdown("*This is an under development project.*")
 st.sidebar.markdown("*Not a replacement for medical advice from a doctor.*")
 
-# Loading message placeholder
-loading_message = st.empty()
-
-def load_model(selected_model_name):
-    loading_message.info("Loading the model, please wait...")
-    model_name = model_links[selected_model_name]
-
-    # Define the BitsAndBytesConfig for quantization
-    bnb_config = BitsAndBytesConfig(
-        load_in_4bit=True,
-        bnb_4bit_quant_type="nf4",
-        bnb_4bit_compute_dtype=torch.bfloat16,
-        bnb_4bit_use_double_quant=False,
-        llm_int8_enable_fp32_cpu_offload=True,
-    )
-
-    # Load the model with quantization settings directly applied
-    model = AutoModelForCausalLM.from_pretrained(
-        model_name,
-        quantization_config=bnb_config,
-        trust_remote_code=True,
-    )
-
-    # Additional configurations and training enhancements
-    model.config.use_cache = False
-    model = prepare_model_for_kbit_training(model)
-
-    # If using PEFT or other enhancements, configure here
-    peft_config = LoraConfig(
-        lora_alpha=16,
-        lora_dropout=0.1,
-        r=64,
-        bias="none",
-        task_type="CAUSAL_LM",
-        target_modules=["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj"],
-    )
-    model = get_peft_model(model, peft_config)
-
-    tokenizer = AutoTokenizer.from_pretrained(
-        "mistralai/Mistral-7B-Instruct-v0.2", trust_remote_code=True
-    )
-
-    # Clear the loading message
-    loading_message.success("Model is ready. Now we are ready!")
-
-    return model, tokenizer
-
-
-# Load model and tokenizer
-model, tokenizer = load_model(selected_model)
-
-# Chat application logic
+
+if "prev_option" not in st.session_state:
+    st.session_state.prev_option = selected_model
+
+if st.session_state.prev_option != selected_model:
+    st.session_state.messages = []
+    # st.write(f"Changed to {selected_model}")
+    st.session_state.prev_option = selected_model
+    reset_conversation()
+
+
+# Pull in the model we want to use
+repo_id = model_links[selected_model]
+
+
+st.subheader(f'AI - {selected_model}')
+# st.title(f'ChatBot Using {selected_model}')
+
+# Set a default model
+if selected_model not in st.session_state:
+    st.session_state[selected_model] = model_links[selected_model]
+
+# Initialize chat history
 if "messages" not in st.session_state:
     st.session_state.messages = []
 
+
+# Display chat messages from history on app rerun
 for message in st.session_state.messages:
     with st.chat_message(message["role"]):
         st.markdown(message["content"])
 
-if prompt := st.chat_input("Ask me anything about diabetes"):
+
+# Accept user input
+if prompt := st.chat_input(f"Hi I'm {selected_model}, ask me a question"):
+
+    # Display user message in chat message container
     with st.chat_message("user"):
         st.markdown(prompt)
-
+    # Add user message to chat history
     st.session_state.messages.append({"role": "user", "content": prompt})
 
-    instructions = """
-    Act as a highly knowledgeable doctor with special interest in diabetes, skilled at explaining complex medical information in a way that is easy to understand for patients without a medical background. Your responses should not only demonstrate empathy and care but also uphold a high standard of medical accuracy and reliability. Respond precisely to what the patient needs in a professional, accurate, and reassuring manner, avoiding any unnecessary information.
-    """
-
-    full_prompt = f"<s>[INST] {prompt} [/INST] {instructions}</s>"
-
+    # Display assistant response in chat message container
     with st.chat_message("assistant"):
-        result = pipeline(
-            task="text-generation",
-            model=model,
-            tokenizer=tokenizer,
-            max_length=1024,
-            temperature=temp_values
-        )(full_prompt)
-
-        generated_text = result[0]['generated_text']
-        response = generated_text.split("</s>")[-1].strip()
-
-        st.markdown(response)
-
-    st.session_state.messages.append({"role": "assistant", "content": response})
+        stream = client.chat.completions.create(
+            model=model_links[selected_model],
+            messages=[
+                {"role": m["role"], "content": m["content"]}
+                for m in st.session_state.messages
+            ],
+            temperature=temp_values,  # 0.5,
+            stream=True,
+            max_tokens=3000,
+        )
+
+        response = st.write_stream(stream)
+    st.session_state.messages.append(
+        {"role": "assistant", "content": response})