transformer: int8 DiT (attn+FFN+modulation, group 64)

text_encoder: int8 Qwen2.5-VL-7B language model (no vision tower)

vae: unquantized AutoencoderKLQwenImage

model_index.json

scheduler: static shift-3 FlowMatchEuler

tokenizer (incl. converted tokenizer.json)

LICENSE (Apache-2.0, additional information per upstream)

NOTICE: NVIDIA OML attribution + statement of quantization changes

README: int8 tier — measured quality, footprints, and why there is no 4-bit tier

initial commit