Add activation learned w6/int8 mixed precision HW quantization of v1 test
#7
by LokkenJP - opened
This PR adds an activation learned w6/int8 mixed precision HW quantization of v1 test.
My usual recipe for a int8 convrot replacement at the same perceived quality but lower space requirements, this time including additional weight savings by quantizing too the Token Refiner tensors.
Actual VRAM usage depends on the workflow and other loaded models.
A full model card and standalone releases will follow only once gating for S3 is lifted.
TenStrip changed pull request status to merged