Fun CN: stream fp8 DiT on CUDA inside @spaces.GPU xlarge (no host bf16)

Fix fp8 stream: bf16 assign then in-place float8 cast + progress

Fun CN: fp8 stream DiT + local bnb4 TE + xlarge (no bf16 host dump / no remote TE)

Deploy NFA Track R FLUX.2 Fun CN ZeroGPU (real depth CN)

Deploy NFA Track R FLUX.2 Fun CN ZeroGPU (real depth CN)

Deploy NFA Track R FLUX.2 Fun CN ZeroGPU (real depth CN)

Deploy NFA Track R FLUX.2 Fun CN ZeroGPU (real depth CN)

Deploy NFA Track R FLUX.2 Fun CN ZeroGPU (real depth CN)

Deploy NFA Track R FLUX.2 Fun CN ZeroGPU (real depth CN)

Deploy NFA Track R FLUX.2 Fun CN ZeroGPU (real depth CN)

Deploy NFA Track R FLUX.2 Fun CN ZeroGPU (real depth CN)

Deploy NFA Track R FLUX.2 Fun CN ZeroGPU (real depth CN)

Deploy NFA Track R FLUX.2 Fun CN ZeroGPU (real depth CN)

initial commit