Spaces:
Sleeping
Sleeping
| from random import randint | |
| from transformers import pipeline, set_seed | |
| import requests | |
| import gradio as gr | |
| import json | |
| # # 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" | |
| title = "text generator based on GPT models" | |
| # TODO TODO TODO TODO support fine tuned models or models for text generation for different purposes | |
| 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"], | |
| ] | |
| # check if api.vicgalle.net:5000/generate is down with timeout of 10 seconds | |
| def is_up(url): | |
| try: | |
| requests.head(url, timeout=10) | |
| return True | |
| except Exception: | |
| return False | |
| # gpt_j_api_down = False | |
| import os | |
| API_URL = "https://api-inference.huggingface.co/models/EleutherAI/gpt-j-6B" | |
| main_gpt_j_api_up = is_up(API_URL) | |
| secondary_gpt_j_api_up = False | |
| if not main_gpt_j_api_up: | |
| # check whether secondary api is available | |
| API_URL = "https://api.vicgalle.net:5000/generate" | |
| secondary_gpt_j_api_up = is_up(API_URL) | |
| headers = {"Authorization": f"Bearer {os.environ['API_TOKEN']}"} | |
| 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: | |
| if main_gpt_j_api_up: | |
| # for this api, a length of > 250 instantly errors, so use a while loop or something | |
| # that would fetch results in chunks of 250 | |
| generated_text = "" | |
| while (max_length > 0): | |
| payload = {"inputs": context, "parameters": {"max_new_tokens": 250, "temperature": temperature, "top_p": top_p}} | |
| response = requests.request("POST", API_URL, data=json.dumps(payload), headers=headers) | |
| context = json.loads(response.content.decode("utf-8"))[0]['generated_text'] | |
| generated_text += context | |
| max_length -= 250 | |
| # payload = {"inputs": context, "parameters":{ | |
| # "max_new_tokens":max_length, "temperature":temperature, "top_p":top_p}} | |
| # data = json.dumps(payload) | |
| # response = requests.request("POST", API_URL, data=data, headers=headers) | |
| # generated_text = json.loads(response.content.decode("utf-8"))[0]['generated_text'] | |
| return generated_text | |
| # use secondary gpt-j-6B api, as the main one is down | |
| if not secondary_gpt_j_api_up: | |
| return "ERR: both GPT-J-6B APIs are down, please try again later (will use a third fallback in the future)" | |
| # use fallback API | |
| # | |
| # 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, | |
| } | |
| response = requests.post( | |
| "http://api.vicgalle.net:5000/generate", params=payload).json() | |
| return response['text'] | |
| else: | |
| # use GPT-2 | |
| # | |
| set_seed(randint(1, 2**31)) | |
| # return sequences specifies how many to return | |
| # for some reson indexing with 'generated-text' doesn't work | |
| # edit: maybe because I was using generated-text, not generated_text (note the underscore in the second) | |
| try: | |
| generated_text = generator(context, max_length=max_length, top_p=top_p, temperature=temperature, num_return_sequences=1)[0]['generated_text'] | |
| except: | |
| generated_text = generator(context, max_length=max_length, top_p=top_p, temperature=temperature, num_return_sequences=1)[0] | |
| return generated_text | |
| # args found in the source: https://github.com/huggingface/transformers/blob/27b3031de2fb8195dec9bc2093e3e70bdb1c4bff/src/transformers/generation_tf_utils.py#L348-L376 | |
| # TODO use fallback gpt-2 inference api for this as well | |
| 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(); | |