Request
Thanks for releasing the StructGen code, dataset, and benchmark!
I am currently comparing multi-reference generation methods and would like to evaluate StructGen under the same setting. The repository documents inference via scripts/run_inference.sh with a local model.safetensors, but I could not find a publicly released checkpoint (Hugging Face models page currently shows 0 models; only the dataset is available).
Could you please release the pretrained StructGen checkpoint(s) used in the paper (e.g., the model.safetensors from the reported training run), ideally on Hugging Face or another public download link?
What would help
- Final inference checkpoint (
model.safetensors) compatible with this repo
- Brief note on the corresponding base model (
BAGEL-7B-MoT) revision / expected directory layout
- Any recommended inference flags for reproducing the paper / benchmark results (e.g.,
placeholder_img_prob, image size, seed)
This would greatly help reproducibility and fair comparison. Thank you!
Request
Thanks for releasing the StructGen code, dataset, and benchmark!
I am currently comparing multi-reference generation methods and would like to evaluate StructGen under the same setting. The repository documents inference via
scripts/run_inference.shwith a localmodel.safetensors, but I could not find a publicly released checkpoint (Hugging Face models page currently shows 0 models; only the dataset is available).Could you please release the pretrained StructGen checkpoint(s) used in the paper (e.g., the
model.safetensorsfrom the reported training run), ideally on Hugging Face or another public download link?What would help
model.safetensors) compatible with this repoBAGEL-7B-MoT) revision / expected directory layoutplaceholder_img_prob, image size, seed)This would greatly help reproducibility and fair comparison. Thank you!