app.py

from transformers import AutoModelForCausalLM, AutoModelForSeq2SeqLM, AutoTokenizer, GenerationConfig, pipeline
import torch
import gradio as gr

# 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")


hf_model_id = "HuggingFaceH4/zephyr-7b-alpha"
model = AutoModelForCausalLM.from_pretrained(hf_model_id)
tokenizerZephyr = AutoTokenizer.from_pretrained(hf_model_id, legacy=False)
generation_config, unused_kwargs = GenerationConfig.from_pretrained(hf_model_id, max_new_tokens=200, temperature=0.7, return_unused_kwargs=True)

model.generation_config = generation_config

pipe = pipeline(
    "text-generation", 
    model=model, 
    tokenizer=tokenizerZephyr,
)
pipe(prompt)


def useZephyr(prompt):
    messages = [
        {
            "role": "system",
            "content": "You are a friendly chatbot who always responds in the style of a pirate.",
        },
        {"role": "user", "content": prompt},
    ]
    # https://huggingface.co/docs/transformers/main/en/chat_templating
    outputs = pipe(prompt)

    return outputs[0]["generated_text"]


def generatePrompt(prompt, max_new_tokens):
    batch = tokenizer(prompt, return_tensors="pt")
    generated_ids = model.generate(
        batch["input_ids"], max_new_tokens=int(max_new_tokens))
    output = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
    prompt = output[0]

    return useZephyr(prompt)


def generate_test(prompt):
    batch = tokenizer(prompt, return_tensors="pt")
    generated_ids = model.generate(batch["input_ids"], max_new_tokens=150)
    output = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
    return output[0]


def generate_prompt(prompt, max_new_tokens):
    return generatePrompt(prompt, max_new_tokens)
#

# Interface


input_prompt = gr.Textbox(label="Prompt", value="photographer")
input_maxtokens = gr.Textbox(label="Max tokens", value="150")
output_component = gr.Textbox(label="Output")
examples = [["photographer"], ["developer"], ["teacher"], [
    "human resources staff"], ["recipe for ham croquettes"]]
description = ""
PerfectGPT = gr.Interface(useZephyr, inputs=[input_prompt, input_maxtokens], outputs=output_component,
                          examples=examples, title="🗿 PerfectGPT v1 🗿", description=description)

PerfectGPT.launch()