Short Review📝
As part of the FLAX/JAX community week organized by 🤗 Hugging Face and the Google Cloud team, they worked on reproducing the results of OpenAI's DALL·E with a smaller architecture. DALL-E can generate new images from any text prompt.
They showed they can achieve impressive results (albeit of a lower quality) while being limited to much smaller hardware resources. Their model is 27 times smaller than the original DALL-E and was trained on a single TPU v3-8 for only 3 days.
Link 🔗
Topics 🤖
#computervision #imagegenerator #texttoimage #deeplearning #VQGAN #encoder #BART #decoder #softmax #crossentropy #CLIP
Modules 📚
#jax #numpy #DalleBart #DalleBartProcessor #vqgan_jax #modeling_flax_vqgan #VQModel #CLIPProcessor #FlaxCLIPModel #transformers #replicate #functools #partial #random #dalle_mini #flax #training #common_utils #shard_prng_key #tqdm #notebook #PIL #Image #trange #shard
Notebook Credit 🌟
For more understanding of the model, refer to the report.