import spaces import torch import torch.nn.functional as F from torch import Tensor from transformers import AutoTokenizer, AutoModel import gradio as gr import os title = """ # 👋🏻Welcome to 🙋🏻‍♂️Tonic's 🐣e5-mistral🛌🏻Embeddings """ description = """ You can use this ZeroGPU Space to test out the current model [intfloat/e5-mistral-7b-instruct](https://huggingface.co/intfloat/e5-mistral-7b-instruct). 🐣e5-mistral🛌🏻 has a larger context🪟window, a different prompting/return🛠️mechanism and generally better results than other embedding models. use it via API to create embeddings or try out the sentence similarity to see how various optimization parameters affect performance. You can also use 🐣e5-mistral🛌🏻 by cloning this space. 🧬🔬🔍 Simply click here: Duplicate Space Join us : 🌟TeamTonic🌟 is always making cool demos! Join our active builder's🛠️community 👻 [![Join us on Discord](https://img.shields.io/discord/1109943800132010065?label=Discord&logo=discord&style=flat-square)](https://discord.gg/GWpVpekp) On 🤗Huggingface: [TeamTonic](https://huggingface.co/TeamTonic) & [MultiTransformer](https://huggingface.co/MultiTransformer) On 🌐Github: [Polytonic](https://github.com/tonic-ai) & contribute to 🌟 [Poly](https://github.com/tonic-ai/poly) 🤗Big thanks to Yuvi Sharma and all the folks at huggingface for the community grant 🤗 """ os.environ['PYTORCH_CUDA_ALLOC_CONF'] = 'max_split_size_mb:30' device = torch.device("cuda" if torch.cuda.is_available() else "cpu") tasks = { 'ArguAna': 'Given a claim, find documents that refute the claim', 'ClimateFEVER': 'Given a claim about climate change, retrieve documents that support or refute the claim', 'DBPedia': 'Given a query, retrieve relevant entity descriptions from DBPedia', 'FEVER': 'Given a claim, retrieve documents that support or refute the claim', 'FiQA2018': 'Given a financial question, retrieve user replies that best answer the question', 'HotpotQA': 'Given a multi-hop question, retrieve documents that can help answer the question', 'MSMARCO': 'Given a web search query, retrieve relevant passages that answer the query', 'NFCorpus': 'Given a question, retrieve relevant documents that best answer the question', 'NQ': 'Given a question, retrieve Wikipedia passages that answer the question', 'QuoraRetrieval': 'Given a question, retrieve questions that are semantically equivalent to the given question', 'SCIDOCS': 'Given a scientific paper title, retrieve paper abstracts that are cited by the given paper', 'SciFact': 'Given a scientific claim, retrieve documents that support or refute the claim', 'Touche2020': 'Given a question, retrieve detailed and persuasive arguments that answer the question', 'TRECCOVID': 'Given a query on COVID-19, retrieve documents that answer the query', } tokenizer = AutoTokenizer.from_pretrained('intfloat/e5-mistral-7b-instruct') model = AutoModel.from_pretrained('intfloat/e5-mistral-7b-instruct', torch_dtype=torch.float16, device_map=device) def last_token_pool(last_hidden_states: Tensor, attention_mask: Tensor) -> Tensor: left_padding = (attention_mask[:, -1].sum() == attention_mask.shape[0]) if left_padding: return last_hidden_states[:, -1] else: sequence_lengths = attention_mask.sum(dim=1) - 1 batch_size = last_hidden_states.shape[0] return last_hidden_states[torch.arange(batch_size, device=last_hidden_states.device), sequence_lengths] class EmbeddingModel: def __init__(self): self.tokenizer = AutoTokenizer.from_pretrained('intfloat/e5-mistral-7b-instruct') self.model = AutoModel.from_pretrained('intfloat/e5-mistral-7b-instruct', torch_dtype=torch.float16, device_map=device) @spaces.GPU def compute_embeddings(selected_task, input_text): max_length = 2042 task_description = tasks[selected_task] processed_texts = [f'Instruct: {task_description}\nQuery: {input_text}'] batch_dict = tokenizer(processed_texts, max_length=max_length - 1, return_attention_mask=False, padding=False, truncation=True) batch_dict['input_ids'] = [input_ids + [tokenizer.eos_token_id] for input_ids in batch_dict['input_ids']] batch_dict = tokenizer.pad(batch_dict, padding=True, return_attention_mask=True, return_tensors='pt') batch_dict = {k: v.to(device) for k, v in batch_dict.items()} outputs = model(**batch_dict) embeddings = last_token_pool(outputs.last_hidden_state, batch_dict['attention_mask']) embeddings = F.normalize(embeddings, p=2, dim=1) embeddings_list = embeddings.detach().cpu().numpy().tolist() return embeddings_list @spaces.GPU def compute_similarity(self, sentence1, sentence2, extra_sentence1, extra_sentence2): # Compute embeddings for each sentence embeddings1 = compute_embeddings(self.selected_task, sentence1) embeddings2 = compute_embeddings(self.selected_task, sentence2) embeddings3 = compute_embeddings(self.selected_task, extra_sentence1) embeddings4 = compute_embeddings(self.selected_task, extra_sentence2) # Convert embeddings to tensors embeddings_tensor1 = torch.tensor(embeddings1).to(device) embeddings_tensor2 = torch.tensor(embeddings2).to(device) embeddings_tensor3 = torch.tensor(embeddings3).to(device) embeddings_tensor4 = torch.tensor(embeddings4).to(device) # Compute cosine similarity similarity1 = F.cosine_similarity(embeddings_tensor1, embeddings_tensor2).item() similarity2 = F.cosine_similarity(embeddings_tensor1, embeddings_tensor3).item() similarity3 = F.cosine_similarity(embeddings_tensor1, embeddings_tensor4).item() return similarity1, similarity2 def app_interface(): with gr.Blocks() as demo: gr.Markdown(title) gr.Markdown(description) with gr.Row(): task_dropdown = gr.Dropdown(list(tasks.keys()), label="Select a Task", value=list(tasks.keys())[0]) with gr.Tab("Embedding Generation"): input_text_box = gr.Textbox(label="📖Input Text") compute_button = gr.Button("Try🐣🛌🏻e5") output_display = gr.Textbox(label="🐣e5-mistral🛌🏻 Embeddings") compute_button.click( fn=EmbeddingModel.compute_embeddings, inputs=[task_dropdown, input_text_box], outputs=output_display ) with gr.Tab("Sentence Similarity"): sentence1_box = gr.Textbox(label="'Focus Sentence' - The 'Subject'") sentence2_box = gr.Textbox(label="'Input Sentence' - 1") extra_sentence1_box = gr.Textbox(label="'Input Sentence' - 2") extra_sentence2_box = gr.Textbox(label="'Input Sentence' - 3") similarity_button = gr.Button("Compute Similarity") similarity_output = gr.Label(label="🐣e5-mistral🛌🏻 Similarity Scores") similarity_button.click( fn=EmbeddingModel.compute_similarity, inputs=[sentence1_box, sentence2_box, extra_sentence1_box, extra_sentence2_box], outputs=similarity_output ) with gr.Row(): with gr.Column(): input_text_box with gr.Column(): compute_button output_display return demo # Run the Gradio app app_interface().launch()