min(DALL·E) is a fast, minimal port of DALL·E Mini to PyTorch
generate_images_stream
and generate_images
to generate individual images. Is in active use in discord bot.temperature
, top_k
, and supercondition_factor
parametersis_reusable
parameter. Turning it off saves memory (e.g. for command line script) and keeping it on makes multiple calls to generate_image
fasterlog2_k
parameter to control top-k image token samplinglog2_supercondition_factor
parameter to control the super conditioning amountlog2_mid_count
and generate_image_stream
to stream intermediate outputs. Incomplete tokens are detokenized to an image multiple times during the decoding process. This adds very little time to the overall run timedtype
parameter to autocast operations to float32
, float16
, or bfloat16
pip install min-dalle
Breaking Changes
MinDalleTorch
is now MinDalle
MinDalleFlax
and flax-to-torch conversion code have been moved to a different repositoryImportant Bug Fixes
New Features
is_expendable
argument reduces memory usage for command line script by loading then unloading encoder/decoder/detokenizer when neededattention_state
replacing 5D keys_values_state
and faster inference timeMinDalleTorch
and MinDalleFlax
classes to initialize model once and run multiple times