Hugging Face's logo

Datasets:
mjuicem
/
tom24-eval-deps

License:
Dataset card Files
xet
 Community
Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 80, in _split_generators
                  raise ValueError(
                  ...<2 lines>...
                  )
              ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

tom24-eval-deps

Extra assets needed to run eval for the postsim_robotwin checkpoint (mjuicem/tom24-postsim-robotwin, step 2000, ToM-World-post-24tasks project). The checkpoint's own transformer weights are not in this repo — see mjuicem/tom24-postsim-robotwin. This repo has everything else the eval pipeline loads at runtime.

Contents

File Size Unpack to Contains
lingbot_va_text.tar 11.4G $TOM_MODELS_ROOT/lingbot-va-base/ text_encoder/, tokenizer/ (T5 text encoder used for language conditioning)
rae_bundle.tar 2.9G $TOM_MODELS_ROOT/ and $RAEV2_ROOT's parent RAEv2-models/ (DINOv3-L encoder + ViT-XL decoder + latent stats, incl. encoders/dinov3/*.pth), RAEv2/ (RAEv2 source, e.g. src/stage1/rae.py, configs/decoder/ViTXL)
vggt_siglip_bundle.tar 14.6G $TOM_MODELS_ROOT/ and vggt repo parent siglip2-so400m-patch14-384/ (SigLIP2 vision-language model), VGGT-Omega/vggt_omega_1b_512.pt, vggt-omega/ (VGGT-Omega source)
eval_misc.tar 4.2M $POSTSIM_DATA/ norm_stats.json (action normalization stats), empty_emb.pt (empty-instruction text embedding fallback)

MD5SUMS has checksums for all four tars — verify after download.

How the paths are wired

These map to the env vars in ToM-World-post-24tasks/config/local.env:

export TOM_MODELS_ROOT=/data/ruanshouwei/Checkpoint     # lingbot-va-base, RAEv2-models, siglip2-so400m-patch14-384, VGGT-Omega
export RAEV2_ROOT=/data/ruanshouwei/Code/tom-wm-rae/RAEv2
export VGGT_REPO=/data/ruanshouwei/Code/tom-wm-lingbot/vggt-omega
export RAE_WEIGHTS_ROOT=$TOM_MODELS_ROOT/RAEv2-models
export POSTSIM_T5_BASE=$TOM_MODELS_ROOT/lingbot-va-base
export POSTSIM_DATA=<your dataset dir>/tom_wm_sim_24tasks_lerobot

Unpack each tar under the directory implied by these vars, e.g.:

mkdir -p "$TOM_MODELS_ROOT" "$(dirname "$RAEV2_ROOT")" "$(dirname "$VGGT_REPO")" "$POSTSIM_DATA"
tar -xf lingbot_va_text.tar     -C "$TOM_MODELS_ROOT/lingbot-va-base"   # creates text_encoder/, tokenizer/
tar -xf rae_bundle.tar          -C "$TOM_MODELS_ROOT"                  # creates RAEv2-models/
tar -xf rae_bundle.tar          -C "$(dirname "$RAEV2_ROOT")"          # also creates RAEv2/ (source)
tar -xf vggt_siglip_bundle.tar  -C "$TOM_MODELS_ROOT"                  # creates siglip2-so400m-patch14-384/, VGGT-Omega/
tar -xf vggt_siglip_bundle.tar  -C "$(dirname "$VGGT_REPO")"           # also creates vggt-omega/ (source)
tar -xf eval_misc.tar           -C "$POSTSIM_DATA"

(rae_bundle.tar and vggt_siglip_bundle.tar each bundle a weights dir plus a source-code dir under different parent paths — extracting twice to two -C targets is intentional and cheap since tar -x just skips paths that don't apply.)

Note: RAEv2/pretrained_models in the original source tree is a symlink to RAEv2-models — it was excluded from rae_bundle.tar to avoid duplicating those 2.7G of weights. After unpacking both pieces, recreate it if the code expects it:

ln -s "$RAE_WEIGHTS_ROOT" "$RAEV2_ROOT/pretrained_models"

Also needed

  • The checkpoint's transformer weights: mjuicem/tom24-postsim-robotwin (diffusion_pytorch_model.safetensors + config.json)
  • Python env: requirements.txt in that same repo (exact pip freeze from the training run)
  • lingbot-va-base/transformer and lingbot-va-base/vae are not included here — eval uses the fine-tuned checkpoint's own transformer, not the base model's.
Downloads last month
48
Total file size:
28.9 GB