app.py

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
from transformers import pipeline
import torch
import gradio as gr
from huggingface_hub import InferenceClient

client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")

# chatgpt-gpt4-prompts-bart-large-cnn-samsum
tokenizer = AutoTokenizer.from_pretrained(
    "Kaludi/chatgpt-gpt4-prompts-bart-large-cnn-samsum")
model = AutoModelForSeq2SeqLM.from_pretrained(
    "Kaludi/chatgpt-gpt4-prompts-bart-large-cnn-samsum", from_tf=True)

# zephyr
    # pipe = pipeline("text-generation", model="HuggingFaceH4/zephyr-7b-alpha",torch_dtype=torch.bfloat16, device_map="auto")

def generateZep(inputuno, inputdos):
    prompt = inputuno
    promptdos = inputdos
    
    generate_kwargs = dict(
        temperature=0.9,
        max_new_tokens=3556,
        top_p=float(0.95),
        repetition_penalty=1.0,
        do_sample=True,
        seed=42,
    )

    batch = tokenizer(prompt, return_tensors="pt")
    generated_ids = model.generate(batch["input_ids"])
    output = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
    new_prompt = output[0]
    
    messages = [
        {
            "role": "system", "content": str(new_prompt)
        },
        {
            "role": "user", "content": str(promptdos)
        },
    ]
    
    stream = client.text_generation(messages, **generate_kwargs, stream=True, details=True, return_full_text=False)
    output = ""

    for response in stream:
        output += response.token.text
        yield output
    return output
    
#

# Interface


input_prompt = gr.Textbox(label="Actua como: ", value="Chef")
input_promptdos = gr.Textbox(label="Prompt: ", value="Recipe for ham croquettes")
output_component = gr.Textbox(label="Output: ")
examples = [["photographer"], ["developer"], ["teacher"], [
    "human resources staff"], ["recipe for ham croquettes"]]
description = ""

PerfectGPT = gr.Interface(fn=generateZep, inputs=[input_prompt, input_promptdos], outputs=output_component, examples=examples, title="🗿 PerfectGPT v1 🗿", description=description)

PerfectGPT.launch()