13 Commits

Author	SHA1	Message	Date
Ryan Dick	823c663e1b	WIP on moving from diffusers to FLUX	2024-08-16 20:22:49 +00:00
Ryan Dick	373b46867a	LLM.int8() quantization is working, but still some rough edges to solve.	2024-08-15 19:34:34 +00:00
Ryan Dick	dc66952491	Clean up NF4 implementation.	2024-08-15 16:30:47 +00:00
Ryan Dick	1b80832b22	NF4 inference working	2024-08-14 23:30:53 +00:00
Ryan Dick	f0baf880b5	Split a FluxTextEncoderInvocation out from the FluxTextToImageInvocation. This has the advantage that we benfit from automatic caching when the prompt isn't changed.	2024-08-12 18:23:02 +00:00
Ryan Dick	a8a2fc106d	Make quantized loading fast for both T5XXL and FLUX transformer.	2024-08-09 19:54:09 +00:00
Ryan Dick	1c97360f9f	Make float16 inference work with FLUX on 24GB GPU.	2024-08-08 18:12:04 -04:00
Ryan Dick	74d6fceeb6	Add support for 8-bit quantizatino of the FLUX T5XXL text encoder.	2024-08-08 18:23:20 +00:00
Ryan Dick	766ddc18dc	Make 8-bit quantization save/reload work for the FLUX transformer. Reload is still very slow with the current optimum.quanto implementation.	2024-08-08 16:40:11 +00:00
Ryan Dick	e6ff7488a1	Minor improvements to FLUX workflow.	2024-08-07 22:10:09 +00:00
Ryan Dick	89a652cfcd	Got FLUX schnell working with 8-bit quantization. Still lots of rough edges to clean up.	2024-08-07 19:50:03 +00:00
Ryan Dick	b227b9059d	Use the FluxPipeline.encode_prompt() api rather than trying to run the two text encoders separately.	2024-08-07 15:12:01 +00:00
Ryan Dick	5dd619e137	First draft of FluxTextToImageInvocation.	2024-08-06 21:51:22 +00:00