from typing import List import numpy as np import requests import gradio as gr import time import os from huggingface_hub import ( create_repo, get_full_repo_name, upload_file, ) class SpaceBuilder: error_message = None url = None @classmethod def split_space_names(cls, names: str) -> List[str]: """ Splits and filters the given space_names. :param names: space names :return: Name List """ name_list = names.split("\n") filtered_list = [] for name in name_list: if not (name == "" or name.isspace()): name = name.replace(" ", "") filtered_list.append(name) return filtered_list @classmethod def file_as_a_string(cls, name_list: List[str], title: str, description: str) -> str: """ Returns the file that is going to be created in the new space as string. :param name_list: list of space names :param title: title :param description: description :return: file as a string """ return ( f"import gradio as gr" f"\nname_list = {name_list}" f"\ninterfaces = [gr.Interface.load(name) for name in name_list]" f"\ngr.mix.Parallel(*interfaces, title=\"{title}\", description=\"{description}\").launch()" ) @classmethod def control_input_and_output_types( cls, interface_list: List["gr.Interface"] ) -> bool: """ Controls whether if input and output types of the given interfaces are the same. :param interface_list: list of interfaces :return: True if all input and output types are the same """ first_input_types = [ type(input) for input in interface_list[0].input_components ] first_output_types = [ type(output) for output in interface_list[0].output_components ] for interface in interface_list: interface_input_types = [ type(input) for input in interface.input_components ] if not np.all( interface_input_types == first_input_types ): # Vectorize the comparison and don't use double for loop cls.error_message = "Provided space input types are different" return False interface_output_types = [ type(output) for output in interface.output_components ] if not np.all(interface_output_types == first_output_types): cls.error_message = "Provided space output types are different" return False return True @classmethod def check_space_name_availability(cls, hf_token: str, space_name: str) -> bool: """ Check whether if the space_name is currently used. :param hf_token: hugging_face token :param space_name: :return: True if the space_name is available """ try: repo_name = get_full_repo_name(model_id=space_name, token=hf_token) except Exception as ex: print(ex) cls.error_message = "You have given an incorrect HuggingFace token" return False try: url = f"https://huggingface.co/spaces/{repo_name}" response = requests.get(url) if response.status_code == 200: cls.error_message = f"The {repo_name} is already used." return False else: print(f"The space name {repo_name} is available") return True except Exception as ex: print(ex) cls.error_message = "Can not send a request to https://huggingface.co" return False @classmethod def load_and_check_spaces(cls, names: str) -> bool: """ Loads given space inputs as interfaces and checks whether if they are loadable. :param names: Input space names :return: True if check is successful """ name_list = cls.split_space_names(names) try: # We could gather these interfaces in parallel if gradio was supporting async gathering. It will probably be possible after the migration to the FastAPI is completed. interfaces = [gr.Interface.load(name) for name in name_list] except Exception as ex: print(ex) cls.error_message = ( f"One of the given space cannot be loaded to gradio, sorry for the inconvenience. " f"\nPlease use different input space names!" ) return False if not cls.control_input_and_output_types(interfaces): return False else: print("Loaded and checked input spaces, great it works!") return True @classmethod def create_space(cls, input_space_names: str, target_space_name: str, hf_token: str, title: str, description: str) -> bool: """ Creates the target space with the given space names. :param input_space_names: Input space name_list :param target_space_name: Target space_name :param hf_token: HuggingFace Write Token :param title: Target Interface Title :param description: Target Interface Description :return: True if success """ name_list = cls.split_space_names(input_space_names) try: create_repo(name=target_space_name, token=hf_token, repo_type="space", space_sdk="gradio") except Exception as ex: print(ex) cls.error_message = "Please provide a correct space name as Only regular characters and '-', '_', '.' accepted. '--' and '..' are forbidden. '-' and '.' cannot start or end the name." return False repo_name = get_full_repo_name(model_id=target_space_name, token=hf_token) try: file_string = cls.file_as_a_string(name_list, title, description) temp_file = open("temp_file.txt", "w") temp_file.write(file_string) temp_file.close() except Exception as ex: print(ex) cls.error_message = "An exception occurred during temporary file writing" return False try: file_url = upload_file( path_or_fileobj="temp_file.txt", path_in_repo="app.py", repo_id=repo_name, repo_type="space", token=hf_token, ) cls.url = f"https://huggingface.co/spaces/{repo_name}" return True except Exception as ex: print(ex) cls.error_message = ( "An exception occurred during writing app.py to the target space" ) return False @staticmethod def build_space( model_or_space_names: str, hf_token: str, target_space_name: str, interface_title: str, interface_description: str ) -> str: """ Creates a space with given input spaces. :param model_or_space_names: Multiple model or space names split with new lines :param hf_token: HuggingFace token :param target_space_name: Target Space Name :param interface_title: Target Interface Title :param interface_description: Target Interface Description :return: """ if ( model_or_space_names== "" or model_or_space_names.isspace() or target_space_name == "" or target_space_name.isspace() or interface_title == "" or interface_title.isspace() or interface_description == "" or interface_description.isspace() ): return "Please fill all the inputs" if hf_token == "" or hf_token.isspace(): hf_token = os.environ['HF_SELF_TOKEN'] if not SpaceBuilder.check_space_name_availability(hf_token=hf_token, space_name=target_space_name): return SpaceBuilder.error_message if not SpaceBuilder.load_and_check_spaces(names=model_or_space_names): return SpaceBuilder.error_message if not SpaceBuilder.create_space(input_space_names=model_or_space_names, target_space_name=target_space_name, hf_token=hf_token, title=interface_title, description=interface_description): return SpaceBuilder.error_message url = SpaceBuilder.url return f"{url}" if __name__ == "__main__": print(f"Gradio Version: {gr.__version__}") iface = gr.Interface( fn=SpaceBuilder.build_space, inputs=[ gr.inputs.Textbox( lines=4, placeholder=( f"Drop model and space links at each line and I will create a new space comparing them. Usage examples:" f"\nspaces/onnx/GPT-2" f"\nmodels/gpt2-large" f"\nmodels/gpt2" ), ), gr.inputs.Textbox(lines=1, placeholder="HuggingFace Write Token"), gr.inputs.Textbox(lines=1, placeholder="Name for the target space, ie. space-building-space"), gr.inputs.Textbox(lines=1, placeholder="Title for the target space interface, ie. Title"), gr.inputs.Textbox(lines=1, placeholder="Description for the target space interface, ie. Description"), ], title="Model Comparator Space Builder", description="Welcome onboard 🤗, I can create a comparative space which will compare the models and/or spaces you provide to me. You can get your HF Write Token from [here](https://huggingface.co/settings/tokens). If you leave HF Token input empty, the space will release under the author's account, [farukozderim](https://huggingface.co/farukozderim). Finally, you can publish spaces as a clone of other spaces if you provide just a single model or space. Have fun :)", outputs=gr.outputs.HTML(label="URL"), examples= [ ["spaces/onnx/GPT-2 \nmodels/gpt2-large \nmodels/EleutherAI/gpt-j-6B", "", "comparison-space", "example-title", "example-description"], ["spaces/onnx/GPT-2", "", "duplicate-space", "example-title", "example-description"], ["models/EleutherAI/gpt-j-6B", "", "space-from-a-model", "example-title", "example-description"] ], ) iface.launch()