Instructions for Setting Up in ComfyUi
Thanks for making the Trellis2-GGUF available.
However some instructions for setting up in ComfyUi would be nice.
In particular where we can find the Custom Node [Aero-Ex/ComfyUI-Trellis2-GGUF]
Also which files we need to download if we want to run say the slat_flow_imgshape2tex_dit_1_3B_512_bf16_Q5_K_M.gguf
And which folders to put them.
Presume we don't need all the files in all the folders.
Well, I've tried my best to get this working on an RTX 2070 Super. The bulk of the early stages can be altered to work with an older kernel under sdpa, but you'll run into kernel limitations in the operations around the 75% mark as the mesh is decoded (which is baked into the main trellis model itself). If you try offloading the process to the CPU, you'll get a memory allocation error. It is a cool project; however, it won't work on older cards.
SPARSE] flex_gemm algorithm 'implicit_gemm_splitk' is not compatible with compute capability 7.5; falling back to 'explicit_gemm'.
RuntimeError: [enforce fail at alloc_cpu.cpp:121] data. DefaultCPUAllocator: not enough memory: you tried to allocate 206158430208 bytes.
Thanks for making the Trellis2-GGUF available.
However some instructions for setting up in ComfyUi would be nice.
In particular where we can find the Custom Node [Aero-Ex/ComfyUI-Trellis2-GGUF]
Also which files we need to download if we want to run say the slat_flow_imgshape2tex_dit_1_3B_512_bf16_Q5_K_M.gguf
And which folders to put them.
Presume we don't need all the files in all the folders.
https://github.com/Aero-Ex/ComfyUI-Trellis2-GGUF/tree/main/trellis2_gguf/modules
make sure this folder is in the right place. node.py has the nodes themselves
However, where should I drop the GGUF models? I don't use huggingface-hub with its constant headache of uploading anything and everything.