---
dataset_info:
- config_name: conv_dataset
  features:
  - name: image
    dtype: image
  - name: dialogs
    list:
    - name: user
      dtype: string
    - name: assistant
      dtype: string
  splits:
  - name: train
    num_bytes: 16058731352.0
    num_examples: 171783
  download_size: 15797887487
  dataset_size: 16058731352.0
- config_name: conv_dataset_abstract
  features:
  - name: image
    dtype: image
  - name: dialogs
    list:
    - name: user
      dtype: string
    - name: assistant
      dtype: string
  splits:
  - name: train
    num_bytes: 12953167683.410126
    num_examples: 139180
  - name: val
    num_bytes: 1619099426.557594
    num_examples: 17397
  - name: test
    num_bytes: 1619285562.0322802
    num_examples: 17399
  download_size: 15997661310
  dataset_size: 16191552672.0
- config_name: conv_dataset_splits
  features:
  - name: image
    dtype: image
  - name: dialogs
    list:
    - name: user
      dtype: string
    - name: assistant
      dtype: string
  splits:
  - name: train
    num_bytes: 12846947688.537003
    num_examples: 137426
  - name: val
    num_bytes: 1605845090.4027524
    num_examples: 17178
  - name: test
    num_bytes: 1605938573.0602446
    num_examples: 17179
  download_size: 15812466569
  dataset_size: 16058731352.0
- config_name: ic_dataset
  features:
  - name: image
    dtype: image
  - name: dialogs
    list:
    - name: user
      dtype: string
    - name: assistant
      dtype: string
  splits:
  - name: train
    num_bytes: 16123260316.0
    num_examples: 173645
  download_size: 15947647876
  dataset_size: 16123260316.0
- config_name: ic_dataset_abstract
  features:
  - name: image
    dtype: image
  - name: dialogs
    list:
    - name: user
      dtype: string
    - name: assistant
      dtype: string
  splits:
  - name: train
    num_bytes: 12898609690.4
    num_examples: 138916
  - name: val
    num_bytes: 1612279785.3674567
    num_examples: 17364
  - name: test
    num_bytes: 1612372637.2325435
    num_examples: 17365
  download_size: 15959794481
  dataset_size: 16123262113.0
configs:
- config_name: conv_dataset
  data_files:
  - split: train
    path: conv_dataset/train-*
- config_name: conv_dataset_abstract
  data_files:
  - split: train
    path: conv_dataset_abstract/train-*
  - split: val
    path: conv_dataset_abstract/val-*
  - split: test
    path: conv_dataset_abstract/test-*
- config_name: conv_dataset_splits
  data_files:
  - split: train
    path: conv_dataset_splits/train-*
  - split: val
    path: conv_dataset_splits/val-*
  - split: test
    path: conv_dataset_splits/test-*
- config_name: ic_dataset
  data_files:
  - split: train
    path: ic_dataset/train-*
- config_name: ic_dataset_abstract
  data_files:
  - split: train
    path: ic_dataset_abstract/train-*
  - split: val
    path: ic_dataset_abstract/val-*
  - split: test
    path: ic_dataset_abstract/test-*
---

### Dataset Summary

This dataset is designed for instruction tuning **vision-language models** (VLMs) on 3D Computer-Aided Design (CAD) comprehension and interactive engineering reasoning. It scales up the [**Text2CAD dataset**](https://arxiv.org/abs/2409.17106) dataset by transforming static CAD assets and multi-level design prompts into a multimodal, conversational format. 

To bridge the gap between static 3D CAD data and conversational AI, we extended the original dataset by:
1. Rendering **multi-view images** for each CAD asset to provide comprehensive visual context.
2. Generating **multi-turn conversations** [multi-turn dialogue] centered around the visual and structural properties of the models, making it ideal for instruction tuning.