Instructions to use cadena45564/cadena with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use cadena45564/cadena with Transformers:
# Use a pipeline as a high-level helper # Warning: Pipeline type "image-to-text" is no longer supported in transformers v5. # You must load the model directly (see below) or downgrade to v4.x with: # 'pip install "transformers<5.0.0' from transformers import pipeline pipe = pipeline("image-to-text", model="cadena45564/cadena")# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("cadena45564/cadena", device_map="auto") - Notebooks
- Google Colab
- Kaggle
CADENA (anonymous copy for review)
Checkpoints accompanying an anonymous ICLR 2027 submission, provided for double-blind review only. CADENA reconstructs a 3D mesh as a parametric CAD program, growing the operation sequence one step at a time and comparing the target with the currently built geometry before choosing the next operation.
- Code: supplementary material of the submission
- Benchmark: https://huggingface.co/datasets/cadena45564/cadena-bench
Checkpoints
| Subfolder | Stage |
|---|---|
sft |
Supervised, final checkpoint of the second stage (CADENA-SFT) |
rl |
Reinforcement learning against executed geometry (CADENA-RL) |
Both are Qwen2-VL-2B policies, 4.4 GB each in bfloat16.
hf download cadena45564/cadena --include 'rl/*' --local-dir ./ckpt
Inference is stepwise and needs the DSL runtime from the code release:
MODEL_PATH=./ckpt/rl DATASET_PATH=data/stepwise_hf/meshes ./inference/run.sh
License
Provided for the purpose of reviewing the submission.