app.py

from random import randint
from transformers import pipeline, set_seed
import requests
import json
import gradio as gr
# # from transformers import AutoModelForCausalLM, AutoTokenizer

# stage, commit, push

# # prompt = "In a shocking finding, scientists discovered a herd of unicorns living in a remote, " \
# #          "previously unexplored valley, in the Andes Mountains. Even more surprising to the " \
# #          "researchers was the fact that the unicorns spoke perfect English."

# ex=None
# try:
#     from transformers import AutoModelForCausalLM, AutoTokenizer
#     tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-j-6B")

#     # "EluttherAI" on this line and for the next occurence only
#     # tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-j-6B")
#     # model = AutoModelForCausalLM.from_pretrained("EleutherAI/gpt-j-6B")
# except Exception as e:
#     ex = e

temperature = gr.inputs.Slider(
    minimum=0, maximum=1.5, default=0.8, label="temperature")
top_p = gr.inputs.Slider(minimum=0, maximum=1.0,
                         default=0.9, label="top_p")

# gradio checkbutton

generator = pipeline('text-generation', model='gpt2')


title = "GPT-J-6B/GPT-2 based text generator"

examples = [
    # another machine learning example
    [["For today's homework assignment, please describe the reasons for the US Civil War."],
        0.8, 0.9, 50, "GPT-2"],
    [["In a shocking discovery, scientists have found a herd of unicorns living in a remote, previously unexplored valley, in the Andes Mountains. Even more surprising to the researchers was the fact that the unicorns spoke perfect English."], 0.8, 0.9, 50, "GPT-2"],
    [["The first step in the process of developing a new language is to invent a new word."],
        0.8, 0.9, 50, "GPT-2"],
]


def f(context, temperature, top_p, max_length, model_idx):
    try:

        # maybe try "0" instead or 1, or "1"
        # use GPT-J-6B
        if model_idx == 0:

            # http://api.vicgalle.net:5000/docs#/default/generate_generate_post
            # https://pythonrepo.com/repo/vicgalle-gpt-j-api-python-natural-language-processing

            payload = {
                "context": context,
                "token_max_length": max_length,  # 512,
                "temperature": temperature,
                "top_p": top_p,
            }
            
            payload = json.dumps(payload)
            response = requests.post(
                "http://api.vicgalle.net:5000/generate", params=payload).json()
            return response['text']
        else:
            # use GPT-2
            # # could easily use the inference API in /gptinference.py but don't know if it supports length>250
            set_seed(randint(1, 2**31))
            # return sequences specifies how many to return
            response = generator(context, max_length=max_length, top_p=top_p,
                             temperature=temperature, num_return_sequences=1)
            print(response)
            return response  # ['generated_text']

            # args found in the source: https://github.com/huggingface/transformers/blob/27b3031de2fb8195dec9bc2093e3e70bdb1c4bff/src/transformers/generation_tf_utils.py#L348-L376

    except Exception as e:
        return f"error with idx{model_idx} : \n"+str(e)


iface = gr.Interface(f, [
    "text",
    temperature,
    top_p,
    gr.inputs.Slider(
        minimum=20, maximum=512, default=30, label="max length"),
    gr.inputs.Dropdown(["GPT-J-6B", "GPT-2"], type="index", label="model"),
], outputs="text", title=title, examples=examples)
iface.launch()  # enable_queue=True

# all below works but testing
# import gradio as gr


# gr.Interface.load("huggingface/EleutherAI/gpt-j-6B",
#     inputs=gr.inputs.Textbox(lines=10, label="Input Text"),
#     title=title, examples=examples).launch();