STARFlow T2I checkpoint, converted to safetensors.
Intended to be used in ComfyUI-STARFlow.
Converted with the following script
import torch
from safetensors.torch import save_file
def main(src="starflow_3B_t2i_256x256.pth", dst="starflow_3B_t2i_256x256.safetensors"):
obj = torch.load(src, map_location="cpu")
if isinstance(obj, dict) and "state_dict" in obj:
obj = obj["state_dict"]
if not isinstance(obj, dict):
raise TypeError(f"Expected a dict/state_dict, got: {type(obj)}")
tensor_dict = {k: v for k, v in obj.items() if isinstance(k, str) and torch.is_tensor(v)}
skipped = len(obj) - len(tensor_dict)
if not tensor_dict:
raise ValueError("No tensors found to save.")
save_file(tensor_dict, dst)
print(f"saved: {dst} (tensors: {len(tensor_dict)}, skipped non-tensors: {skipped})")
if __name__ == "__main__":
main()
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
🙋
Ask for provider support