Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,93 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import spaces
|
2 |
+
import torch
|
3 |
+
import torch.nn.functional as F
|
4 |
+
from torch import Tensor
|
5 |
+
from transformers import AutoTokenizer, AutoModel
|
6 |
+
import gradio as gr
|
7 |
+
|
8 |
+
title = """
|
9 |
+
# 👋🏻Welcome to 🙋🏻♂️Tonic's 🐣e5-mistral🛌🏻Embeddings """
|
10 |
+
description = """
|
11 |
+
You can use this Space to test out the current model [intfloat/e5-mistral-7b-instruct](https://huggingface.co/intfloat/e5-mistral-7b-instruct). e5mistral has a larger context window, a different prompting/return mechanism and generally better results than other embedding models.
|
12 |
+
You can also use 🐣e5-mistral🛌🏻 by cloning this space. 🧬🔬🔍 Simply click here: <a style="display:inline-block" href="https://huggingface.co/spaces/Tonic/e5?duplicate=true"><img src="https://img.shields.io/badge/-Duplicate%20Space-blue?labelColor=white&style=flat&logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAAAXNSR0IArs4c6QAAAP5JREFUOE+lk7FqAkEURY+ltunEgFXS2sZGIbXfEPdLlnxJyDdYB62sbbUKpLbVNhyYFzbrrA74YJlh9r079973psed0cvUD4A+4HoCjsA85X0Dfn/RBLBgBDxnQPfAEJgBY+A9gALA4tcbamSzS4xq4FOQAJgCDwV2CPKV8tZAJcAjMMkUe1vX+U+SMhfAJEHasQIWmXNN3abzDwHUrgcRGmYcgKe0bxrblHEB4E/pndMazNpSZGcsZdBlYJcEL9Afo75molJyM2FxmPgmgPqlWNLGfwZGG6UiyEvLzHYDmoPkDDiNm9JR9uboiONcBXrpY1qmgs21x1QwyZcpvxt9NS09PlsPAAAAAElFTkSuQmCC&logoWidth=14" alt="Duplicate Space"></a></h3>
|
13 |
+
Join us : 🌟TeamTonic🌟 is always making cool demos! Join our active builder's🛠️community on 👻Discord: [](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)
|
14 |
+
"""
|
15 |
+
# Define the function to pool the last token
|
16 |
+
def last_token_pool(last_hidden_states: Tensor, attention_mask: Tensor) -> Tensor:
|
17 |
+
left_padding = (attention_mask[:, -1].sum() == attention_mask.shape[0])
|
18 |
+
if left_padding:
|
19 |
+
return last_hidden_states[:, -1]
|
20 |
+
else:
|
21 |
+
sequence_lengths = attention_mask.sum(dim=1) - 1
|
22 |
+
batch_size = last_hidden_states.shape[0]
|
23 |
+
return last_hidden_states[torch.arange(batch_size, device=last_hidden_states.device), sequence_lengths]
|
24 |
+
|
25 |
+
# Define the function to get detailed instruct
|
26 |
+
def get_detailed_instruct(task_description: str, query: str) -> str:
|
27 |
+
return f'Instruct: {task_description}\nQuery: {query}'
|
28 |
+
|
29 |
+
# Load tokenizer and model
|
30 |
+
tokenizer = AutoTokenizer.from_pretrained('intfloat/e5-mistral-7b-instruct')
|
31 |
+
model = AutoModel.from_pretrained('intfloat/e5-mistral-7b-instruct')
|
32 |
+
|
33 |
+
@spaces.GPU
|
34 |
+
def compute_embeddings(*input_texts):
|
35 |
+
# Check if GPU is available and use it; otherwise, use CPU
|
36 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
37 |
+
|
38 |
+
# Move model to the chosen device
|
39 |
+
model.to(device)
|
40 |
+
max_length = 4096
|
41 |
+
task = 'Given a web search query, retrieve relevant passages that answer the query'
|
42 |
+
|
43 |
+
# Prepare the input texts
|
44 |
+
processed_texts = [get_detailed_instruct(task, text) for text in input_texts]
|
45 |
+
|
46 |
+
# Tokenize the input texts
|
47 |
+
batch_dict = tokenizer(processed_texts, max_length=max_length - 1, return_attention_mask=False, padding=False, truncation=True)
|
48 |
+
batch_dict['input_ids'] = [input_ids + [tokenizer.eos_token_id] for input_ids in batch_dict['input_ids']]
|
49 |
+
batch_dict = tokenizer.pad(batch_dict, padding=True, return_attention_mask=True, return_tensors='pt')
|
50 |
+
|
51 |
+
# Get model outputs
|
52 |
+
outputs = model(**batch_dict)
|
53 |
+
embeddings = last_token_pool(outputs.last_hidden_state, batch_dict['attention_mask'])
|
54 |
+
|
55 |
+
# Normalize embeddings
|
56 |
+
embeddings = F.normalize(embeddings, p=2, dim=1)
|
57 |
+
return embeddings.detach().cpu().numpy()
|
58 |
+
|
59 |
+
|
60 |
+
def app_interface():
|
61 |
+
with gr.Blocks() as demo:
|
62 |
+
gr.Markdown(title)
|
63 |
+
gr.Markdown(description)
|
64 |
+
|
65 |
+
# Input text boxes
|
66 |
+
input_text_boxes = [gr.Textbox(label=f"Input Text {i+1}") for i in range(4)]
|
67 |
+
|
68 |
+
# Button to compute embeddings
|
69 |
+
compute_button = gr.Button("Compute Embeddings")
|
70 |
+
|
71 |
+
# Output display
|
72 |
+
output_display = gr.Dataframe(headers=["Embedding"], datatype=["numpy"])
|
73 |
+
|
74 |
+
# Layout
|
75 |
+
with gr.Row():
|
76 |
+
with gr.Column():
|
77 |
+
for text_box in input_text_boxes:
|
78 |
+
text_box.render()
|
79 |
+
with gr.Column():
|
80 |
+
compute_button.render()
|
81 |
+
output_display.render()
|
82 |
+
|
83 |
+
# Function call
|
84 |
+
compute_button.click(
|
85 |
+
fn=compute_embeddings,
|
86 |
+
inputs=input_text_boxes,
|
87 |
+
outputs=output_display
|
88 |
+
)
|
89 |
+
|
90 |
+
return demo
|
91 |
+
|
92 |
+
# Run the Gradio app
|
93 |
+
app_interface().launch()
|