app.py

import gradio as gr
from huggingface_hub import InferenceClient

# Function to return the appropriate client based on the model selected
def client_fn(model):
 model_map = {
 "Nous Hermes Mixtral 8x7B DPO": "NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO",
 "StarChat2 15b": "HuggingFaceH4/starchat2-15b-v0.1",
 "Mistral 7B v0.3": "mistralai/Mistral-7B-Instruct-v0.3",
 "Phi 3 mini": "microsoft/Phi-3-mini-4k-instruct",
 "Mixtral 8x7B": "mistralai/Mixtral-8x7B-Instruct-v0.1"
 }
 return InferenceClient(model_map.get(model, "mistralai/Mixtral-8x7B-Instruct-v0.1"))

system_instructions = ("[SYSTEM] You are a chat bot named 'H go'."
 "Your task is to Answer the question."
 "Keep conversation very short, clear and concise."
 "Respond naturally and concisely to the user's queries. "
 "The expectation is that you will avoid introductions and start answering the query directly, Only answer the question asked by user, Do not say unnecessary things."
 "Begin with a greeting if the user initiates the conversation. "
 "Here is the user's query:[QUESTION] ")

# Function to generate model responses
def models(text, model="Mixtral 8x7B"):
 client = client_fn(model)
 generate_kwargs = {
 "max_new_tokens": 100,
 "do_sample": True,
 }
 
 formatted_prompt = f"{system_instructions} {text} [ANSWER]"
 stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
 
 output = ""
 for response in stream:
 output += response.token.text
 if output.endswith("</s>"):
 output = output[:-4]
 return output

# Gradio interface description and configuration
description = """# H GO
### Inspired from Google Go"""

with gr.Blocks() as demo:
 gr.Markdown(description)
 
 text_input = gr.Textbox(label="Enter your message here:")
 dropdown = gr.Dropdown(['Mixtral 8x7B', 'Nous Hermes Mixtral 8x7B DPO', 'StarChat2 15b', 'Mistral 7B v0.3', 'Phi 3 mini'], value="Mistral 7B v0.3", label="Select Model")
 submit_btn = gr.Button("Send")
 output_text = gr.Textbox(label="Response")
 
 submit_btn.click(fn=models, inputs=[text_input, dropdown], outputs=output_text)

# Queue and launch configuration for Gradio
demo.queue(max_size=300000)
demo.launch()