/flux-fp8-api

Flux diffusion model implementation using quantized fp8 matmul & remaining layers use faster half precision accumulate, which is ~2x faster on consumer devices.

Primary LanguagePythonApache License 2.0Apache-2.0

Watchers