import gradio as gr
import spaces

import transformers
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "doubledsbv/Llama-3-Kafka-8B-v0.3"

model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16)
tokenizer = transformers.AutoTokenizer.from_pretrained(model_id)


generate_text = transformers.pipeline(
    model=model, tokenizer=tokenizer,
    return_full_text=True,  
    task='text-generation',
    device="gpu",
)

@spaces.GPU
def chat_function(message, history, system_prompt,max_new_tokens,temperature):
    messages = [
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": message},
    ]
    prompt = pipeline.tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True
    )
    terminators = [
        pipeline.tokenizer.eos_token_id,
        pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")
    ]
    outputs = pipeline(
        prompt,
        max_new_tokens=max_new_tokens,
        eos_token_id=terminators,
        do_sample=True,
        temperature=temperature,
        top_p=0.9,
    )
    return outputs[0]["generated_text"][len(prompt):]

gr.ChatInterface(
    chat_function,
    chatbot=gr.Chatbot(height=400),
    textbox=gr.Textbox(placeholder="Enter message here", container=False, scale=7),
    title="Llama-3-Kafka-8B-v0.3",
    description="""
    German-focused finetuned version of Llama-3-8B
    """,
    additional_inputs=[
        gr.Textbox("Du bist ein freundlicher KI-Assistent", label="System Prompt"),
        gr.Slider(512, 8192, label="Max New Tokens"),
        gr.Slider(0, 1, label="Temperature")
    ]
).launch()