app.py

import gradio as gr
import spaces
import torch
from transformers import AutoTokenizer, AutoModel
from sklearn.decomposition import PCA
import plotly.graph_objects as go
from huggingface_hub import HfApi
from huggingface_hub import hf_hub_download
import os
import sys

model_name = "sentence-transformers/all-MiniLM-L6-v2"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModel.from_pretrained(model_name)

@spaces.GPU
def get_embedding(text):
    inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=512)
    with torch.no_grad():
        outputs = model(**inputs)
    return outputs.last_hidden_state.mean(dim=1).squeeze().numpy()

def compress_to_3d(embedding):
    pca = PCA(n_components=3)
    return pca.fit_transform(embedding.reshape(1, -1))[0]

@spaces.GPU
def compare_embeddings(text1, text2):
    emb1 = get_embedding(text1)
    emb2 = get_embedding(text2)
    
    emb1_3d = compress_to_3d(emb1)
    emb2_3d = compress_to_3d(emb2)
    
    fig = go.Figure(data=[
        go.Scatter3d(x=[0, emb1_3d[0]], y=[0, emb1_3d[1]], z=[0, emb1_3d[2]], mode='lines+markers', name='Text 1'),
        go.Scatter3d(x=[0, emb2_3d[0]], y=[0, emb2_3d[1]], z=[0, emb2_3d[2]], mode='lines+markers', name='Text 2')
    ])
    
    fig.update_layout(scene=dict(xaxis_title='X', yaxis_title='Y', zaxis_title='Z'))
    
    return fig

iface = gr.Interface(
    fn=compare_embeddings,
    inputs=[
        gr.Textbox(label="Text 1"),
        gr.Textbox(label="Text 2")
    ],
    outputs=gr.Plot(),
    title="3D Embedding Comparison",
    description="Compare the embeddings of two strings visualized in 3D space."
)

iface.launch()
demo.launch()