import logging import re import textwrap from urllib.request import urlopen import chromadb import gradio as gr import posthog import torch from bs4 import BeautifulSoup from chromadb.config import Settings from sentence_transformers import SentenceTransformer from modules import chat, shared logging.info('Intercepting all calls to posthog :)') posthog.capture = lambda *args, **kwargs: None class Collecter(): def __init__(self): pass def add(self, texts: list[str]): pass def get(self, search_strings: list[str], n_results: int) -> list[str]: pass def clear(self): pass class Embedder(): def __init__(self): pass def embed(self, text: str) -> list[torch.Tensor]: pass class ChromaCollector(Collecter): def __init__(self, embedder: Embedder): super().__init__() self.chroma_client = chromadb.Client(Settings(anonymized_telemetry=False)) self.embedder = embedder self.collection = self.chroma_client.create_collection(name="context", embedding_function=embedder.embed) self.ids = [] def add(self, texts: list[str]): self.ids = [f"id{i}" for i in range(len(texts))] self.collection.add(documents=texts, ids=self.ids) def get(self, search_strings: list[str], n_results: int) -> list[str]: n_results = min(len(self.ids), n_results) result = self.collection.query(query_texts=search_strings, n_results=n_results, include=['documents'])['documents'][0] return result def get_ids(self, search_strings: list[str], n_results: int) -> list[str]: n_results = min(len(self.ids), n_results) result = self.collection.query(query_texts=search_strings, n_results=n_results, include=['documents'])['ids'][0] return list(map(lambda x : int(x[2:]), result)) def clear(self): self.collection.delete(ids=self.ids) class SentenceTransformerEmbedder(Embedder): def __init__(self) -> None: self.model = SentenceTransformer("sentence-transformers/all-mpnet-base-v2") self.embed = self.model.encode embedder = SentenceTransformerEmbedder() collector = ChromaCollector(embedder) chunk_count = 5 def add_chunks_to_collector(chunks): global collector collector.clear() collector.add(chunks) def feed_data_into_collector(corpus, chunk_len): # Defining variables chunk_len = int(chunk_len) cumulative = '' # Breaking the data into chunks and adding those to the db cumulative += "Breaking the input dataset...\n\n" yield cumulative data_chunks = [corpus[i:i + chunk_len] for i in range(0, len(corpus), chunk_len)] cumulative += f"{len(data_chunks)} chunks have been found.\n\nAdding the chunks to the database...\n\n" yield cumulative add_chunks_to_collector(data_chunks) cumulative += "Done." yield cumulative def feed_file_into_collector(file, chunk_len): yield 'Reading the input dataset...\n\n' text = file.decode('utf-8') for i in feed_data_into_collector(text, chunk_len): yield i def feed_url_into_collector(urls, chunk_len): urls = urls.strip().split('\n') all_text = '' cumulative = '' for url in urls: cumulative += f'Loading {url}...\n\n' yield cumulative html = urlopen(url).read() soup = BeautifulSoup(html, features="html.parser") for script in soup(["script", "style"]): script.extract() text = soup.get_text() lines = (line.strip() for line in text.splitlines()) chunks = (phrase.strip() for line in lines for phrase in line.split(" ")) text = '\n\n'.join(chunk for chunk in chunks if chunk) all_text += text for i in feed_data_into_collector(all_text, chunk_len): yield i def apply_settings(_chunk_count): global chunk_count chunk_count = int(_chunk_count) settings_to_display = { 'chunk_count': chunk_count, } yield f"The following settings are now active: {str(settings_to_display)}" def input_modifier(string): if shared.is_chat(): return string # Find the user input pattern = re.compile(r"<\|begin-user-input\|>(.*?)<\|end-user-input\|>", re.DOTALL) match = re.search(pattern, string) if match: user_input = match.group(1).strip() else: user_input = '' # Get the most similar chunks results = collector.get(user_input, n_results=chunk_count) # Make the replacements string = string.replace('<|begin-user-input|>', '') string = string.replace('<|end-user-input|>', '') string = string.replace('<|injection-point|>', '\n'.join(results)) return string def custom_generate_chat_prompt(user_input, state, **kwargs): if len(shared.history['internal']) > 2 and user_input != '': chunks = [] for i in range(len(shared.history['internal'])-1): chunks.append('\n'.join(shared.history['internal'][i])) add_chunks_to_collector(chunks) query = '\n'.join(shared.history['internal'][-1] + [user_input]) try: best_ids = collector.get_ids(query, n_results=len(shared.history['internal'])-1) # Sort the history by relevance instead of by chronological order, # except for the latest message state['history'] = [shared.history['internal'][id_] for id_ in best_ids[::-1]] + [shared.history['internal'][-1]] except RuntimeError: logging.error("Couldn't query the database, moving on...") return chat.generate_chat_prompt(user_input, state, **kwargs) def ui(): with gr.Accordion("Click for more information...", open=False): gr.Markdown(textwrap.dedent(""" ## About This extension takes a dataset as input, breaks it into chunks, and adds the result to a local/offline Chroma database. The database is then queried during inference time to get the excerpts that are closest to your input. The idea is to create an arbitrarily large pseudocontext. It is a modified version of the superbig extension by kaiokendev: https://github.com/kaiokendev/superbig ## Notebook/default modes ### How to use it 1) Paste your input text (of whatever length) into the text box below. 2) Click on "Load data" to feed this text into the Chroma database. 3) In your prompt, enter your question between `<|begin-user-input|>` and `<|end-user-input|>`, and specify the injection point with `<|injection-point|>`. By default, the 5 closest chunks will be injected. You can customize this value in the "Generation settings" tab. The special tokens mentioned above (`<|begin-user-input|>`, `<|end-user-input|>`, and `<|injection-point|>`) are removed when the injection happens. ### Example For your convenience, you can use the following prompt as a starting point (for Alpaca models): ``` Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request. ### Instruction: You are ArxivGPT, trained on millions of Arxiv papers. You always answer the question, even if full context isn't provided to you. The following are snippets from an Arxiv paper. Use the snippets to answer the question. Think about it step by step <|injection-point|> ### Input: <|begin-user-input|> What datasets are mentioned in the paper above? <|end-user-input|> ### Response: ``` ## Chat mode In chat mode, the extension automatically sorts the history by relevance instead of chronologically, except for the very latest input/reply pair. That is, the prompt will include (starting from the end): * Your input * The latest input/reply pair * The #1 most relevant input/reply pair prior to the latest * The #2 most relevant input/reply pair prior to the latest * Etc This way, the bot can have a long term history. *This extension is currently experimental and under development.* """)) if not shared.is_chat(): with gr.Row(): with gr.Column(): with gr.Tab("Text input"): data_input = gr.Textbox(lines=20, label='Input data') update_data = gr.Button('Load data') with gr.Tab("URL input"): url_input = gr.Textbox(lines=10, label='Input URLs', info='Enter one or more URLs separated by newline characters.') update_url = gr.Button('Load data') with gr.Tab("File input"): file_input = gr.File(label='Input file', type='binary') update_file = gr.Button('Load data') with gr.Tab("Generation settings"): chunk_count = gr.Number(value=5, label='Chunk count', info='The number of closest-matching chunks to include in the prompt.') update_settings = gr.Button('Apply changes') chunk_len = gr.Number(value=700, label='Chunk length', info='In characters, not tokens. This value is used when you click on "Load data".') with gr.Column(): last_updated = gr.Markdown() update_data.click(feed_data_into_collector, [data_input, chunk_len], last_updated, show_progress=False) update_url.click(feed_url_into_collector, [url_input, chunk_len], last_updated, show_progress=False) update_file.click(feed_file_into_collector, [file_input, chunk_len], last_updated, show_progress=False) update_settings.click(apply_settings, [chunk_count], last_updated, show_progress=False)