real hound
Hello, I entered the problem with cached model.
I am trying to use cached model with my huggingface model (aggregate multiple safetensors into one model) Helg/WanComfyServerless (https://huggingface.co/Helg/WanComfyServerless).
Model is public and no access token required.
Total size: 141 GB
model file structure:
split_files/
....diffusion_models/
........fun_vace_high.safetensors
........wan2.2_i2v_high_noise_14B_fp16.safetensors
........etc
....loras/
....text_encoders/
....vae/
But when worker ready there is log: image ready, model not found. And it stuck.
Problem is consistent on all workers in different datacenters.
I tried recreate enpoint.
Before I used pods with custom orchestrator and images with baked models and decided to move to serverless to reduce image size and speed up cold starts.
inner valley