zerogpu-aoti/FLUX.1-dev-fa3-aoti

Update README.md

bf16 fa3

Enable aoti

fp8e4m3 (disable aoti)

gr.Gallery()

FLUX.1-dev

fuse_qkv_projections

Revert "qkv"

qkv

fa3 + big refresh

Demo touch-ups

No progress for now

Empty commit

9 images + No pipeline progress

Useless here

Streaming

Touch ups

Further optimize + quantize

No need torchvision

Not needed anymore

Fixes

PyTorch 2.9 issue?

Real file

Cleanup

Alright

ZeroGPUCompiledModel

Timings

Restore transformer config

Duration

Some basic requirements

Begin actual demo

Brainstorming backup

Two calls in a row

Worker re-use

Move weights to CUDA

Fix CUDA path

Package AOTI

detach after copy_

Keep pinned-memory

requires_grad=False

Load constants

Phase 1

gitignore .pyc

PyTorch upgrade + torchvision

--progress=dot:giga

Simpler

wget --progress=dot:mega

f-strings

Test AOTI base example during app startup

CUDA toolkit install script