Downloads · 30 days
0
stage3-epoch0-step100-20251128_220651
omnicad-lab-L3d/stage3-epoch0-step100-20251128_220651
stage3-epoch0-step100-20251128_220651 is a any-to-any model from omnicad-lab-L3d. Use it for the any-to-any task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
The model uses the following token sequence order for multimodal inputs:
Access
Public
Updated Nov 29, 2025
Repo size
1.2 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt822 MB · 69%
At a glance
- Task
- Any-to-Any
- License
- mit
- Access
- Public
- Created
- Nov 29, 2025
- Updated
- Nov 29, 2025
- SHA
- 37a93879
Base models
Model card
From the Hugging Face model README
The model uses the following token sequence order for multimodal inputs:
[Image tokens] [Point Cloud tokens] [Instruction + Description] [CAD Commands]
256 1+256 variable variable
python scripts/train_curriculum.py \
--use_wandb \
--output_dir ./CAD-MLLM-checkpoints \
--omnicad_json_root ./data/Omni-CAD-subset-complete/json \
--omnicad_txt_path ./data/Omni-CAD-subset-complete/txt \
--omnicad_img_root ./data/Omni-CAD-subset-complete/img \
--omnicad_pc_root ./data/Omni-CAD-subset-complete/pointcloud \
--llm_model_name "Qwen/Qwen3-8B" \
--resume_from_ckpt "./CAD-MLLM-checkpoints/stage3_all/checkpoint-epoch0-step100-20251128_220651" \
--start_from_stage 3 \
--stage1_epochs 0 \
--stage2_epochs 0 \
--stage3_epochs 10 \
--stage3_lr 2e-5 \
--max_seq_length 1024 \
--batch_size 1 \
--gradient_accumulation_steps 16 \
--lora_r 32 \
--lora_alpha 64 \
--warmup_steps 50 \
--logging_steps 1000 \
--save_steps 50 \
--device cuda \
--dtype bfloat16 \
--wandb_project "CAD-MLLM-stage_3" \
--wandb_run_name "latent_pc" \
--seed 30
- Task
- Any-to-Any
- License
- mit
- Created
- Nov 29, 2025
- Updated
- Nov 29, 2025