app.py

import gradio as gr
from huggingface_hub import InferenceClient

# Default client with the first model
client = InferenceClient("mistralai/Mistral-7B-Instruct-v0.3")

# Function to switch between models based on selection
def switch_client(model_name: str):
    return InferenceClient(model_name)

# Define presets for each model
presets = {
    "mistralai/Mistral-7B-Instruct-v0.3": {
        "Fast": {"max_tokens": 256, "temperature": 1.0, "top_p": 0.9},
        "Normal": {"max_tokens": 512, "temperature": 0.7, "top_p": 0.95},
        "Quality": {"max_tokens": 1024, "temperature": 0.5, "top_p": 0.90},
        "Unreal Performance": {"max_tokens": 2048, "temperature": 0.6, "top_p": 0.75},
    }
}

# Fixed system message
SYSTEM_MESSAGE = "Your name is Lake 1 Base but my is User"

def respond(
    message,
    history: list,
    model_name,
    preset_name
):
    # Switch client based on model selection
    global client
    client = switch_client(model_name)
    
    messages = [{"role": "system", "content": SYSTEM_MESSAGE}]

    # Ensure history is a list of dictionaries
    for val in history:
        if isinstance(val, dict) and 'role' in val and 'content' in val:
            messages.append({"role": val['role'], "content": val['content']})

    messages.append({"role": "user", "content": message})

    # Get the preset settings
    preset = presets[model_name][preset_name]
    max_tokens = preset["max_tokens"]
    temperature = preset["temperature"]
    top_p = preset["top_p"]

    # Get the response from the model
    response = client.chat_completion(
        messages,
        max_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
    )

    # Extract the content from the response
    final_response = response.choices[0].message['content']
    
    return final_response

# Model names and their pseudonyms
model_choices = [
    ("mistralai/Mistral-7B-Instruct-v0.3", "Lake 1 Base")
]

# Convert pseudonyms to model names for the dropdown
pseudonyms = [model[1] for model in model_choices]

# Function to handle model selection and pseudonyms
def respond_with_pseudonym(
    message,
    history: list,
    selected_model,  # Change this to selected_model
    selected_preset
):
    print(f"Selected Model: {selected_model}")  # Debugging line
    print(f"Available Models: {pseudonyms}")  # Debugging line

    # Find the actual model name from the pseudonym
    try:
        model_name = next(model[0] for model in model_choices if model[1] == selected_model)
    except StopIteration:
        return f"Error: The selected model '{selected_model}' is not valid. Please select a valid model."

    # Call the existing respond function
    response = respond(message, history, model_name, selected_preset)
    
    return response

# Gradio Chat Interface
demo = gr.ChatInterface(
    fn=respond_with_pseudonym,
    additional_inputs=[
        gr.Dropdown(choices=pseudonyms, label="Select Model", value=pseudonyms[0]),  # Pseudonym selection dropdown
        gr.Dropdown(choices=list(presets["mistralai/Mistral-7B-Instruct-v0.3"].keys()), label="Select Preset", value="Fast")  # Preset selection dropdown
        
    ],
)

if __name__ == "__main__":
    demo.launch()