🔤 CS180 HW5: T5 Text Encoder Embeddings Generator

This space uses the DeepFloyd IF T5 text encoder to generate embeddings from your text prompts.

How to use:

Enter your prompts in the text box (one prompt per line)
Click "Generate Embeddings"
Download the generated .pth file containing the embeddings

About the embeddings:

Each embedding has shape: [1, 77, 4096]
77 = max sequence length
4096 = embedding dimension of the T5 encoder
An empty prompt ('') is automatically added for Classifier Free Guidance (CFG)

Enter Prompts (one per line)

Status

Download Embeddings (.pth file)

📝 Note:

The first run may take a while as the model needs to download (~8GB)
Subsequent runs will be faster
The generated .pth file can be loaded in PyTorch using: torch.load('prompt_embeds_dict.pth')