JDWebProgrammer/text-embeddings-transformers

1import gradio as gr 2from transformers import AutoTokenizer, AutoModel3import torch4 5tokenizer = AutoTokenizer.from_pretrained('intfloat/multilingual-e5-large')6model = AutoModel.from_pretrained('intfloat/multilingual-e5-large')7 8def mean_pooling(model_output, attention_mask):9    token_embeddings = model_output[0]10    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()11    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)12 13def encode_sentences(sentences):14    encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')15    with torch.no_grad():16        model_output = model(**encoded_input)17    sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])18    return sentence_embeddings.tolist()19 20demo = gr.Interface(fn=encode_sentences,21                    inputs="textbox",22                    outputs="text")23 24if __name__ == "__main__":25    demo.launch()