Skip to content

SolarRenderer: Few-Step Generative Reference Rendering

Python
745 B
Raw Blame
# Model dependencies resolved with Reactor Runtime during `reactor build`.
# `build.runtime_version` in reactor.yaml selects the Runtime release.

# Native H3 encoder, transformer and causal LoRA loader.
torch==2.8.0
torchvision==0.23.0
diffusers==0.40.0
transformers==5.14.1
peft==0.20.0
accelerate>=1.14
safetensors>=0.8
huggingface-hub>=1.25
fsspec>=2024.6
einops>=0.8
imageio==2.37.4
imageio-ffmpeg==0.6.0

# Four-GPU causal attention. CUTLASS 4.8 lifts the protobuf <7 constraint.
flash-attn-4==4.0.0b34
nvidia-cutlass-dsl==4.8.1
quack-kernels==0.6.5

# CPU media validation and configuration; Runtime shares these dependencies.
numpy>=2
PyYAML>=6
Pillow>=11
av>=18

# Flash Attention builds after torch is available.
ninja
packaging
wheel