tracebloc Model Zoo 🧠
Pre-built ML models you can train and benchmark on the tracebloc platform — or use as a starting point for your own.
Every model in this repo is compatible with tracebloc's secure training environment. Pick one, upload it, and start an experiment in minutes. No setup, no configuration, no boilerplate.
Models
| Task | Framework | Path | Use when you need to… |
|---|
| Image classification | PyTorch | model_zoo/image_classification/pytorch/ | Categorize images into labels |
| Object detection | PyTorch | model_zoo/object_detection/pytorch/ | Locate and classify objects in images |
| Text classification | PyTorch | model_zoo/text_classification/pytorch/ | Classify documents, reviews, tickets |
| Token classification | PyTorch | model_zoo/token_classification/pytorch/ | Tag tokens: NER, POS, slot filling |
| Causal language modeling | PyTorch | model_zoo/causal_language_modeling/pytorch/ | Pretrain/fine-tune a decoder-only LM (next-token) |
| Masked language modeling | PyTorch | model_zoo/masked_language_modeling/pytorch/ | Pretrain/domain-adapt an encoder LM (fill-in-the-blank) |
| Seq2seq | PyTorch | model_zoo/seq2seq/pytorch/ | Map text to text: translation, summarization |
| Embeddings | PyTorch | model_zoo/embeddings/pytorch/ | Learn text embeddings for search, retrieval, similarity |
| Sentence pair classification | PyTorch | model_zoo/sentence_pair_classification/pytorch/ | Score sentence pairs: NLI, duplicates, relevance |
| Semantic segmentation | PyTorch | model_zoo/semantic_segmentation/pytorch/ | Pixel-level image labeling |
| Keypoint detection | PyTorch | model_zoo/keypoint_detection/pytorch/ | Detect landmarks on objects or bodies |
| Tabular classification | PyTorch, sklearn | model_zoo/tabular_classification/ | Classify structured/tabular data |
| Tabular regression | PyTorch, sklearn | model_zoo/tabular_regression/ | Predict continuous values from tables |
| Time series forecasting | PyTorch | model_zoo/time_series_forecasting/pytorch/ | Forecast future values from sequences |
| Time series classification | PyTorch | model_zoo/time_series_classification/pytorch/ | Assign one label to each whole sequence |
| Time-to-event prediction | lifelines, PyTorch, scikit-survival | model_zoo/time_to_event_prediction/ | Predict when an event will occur |
Quick start
git clone https://github.com/tracebloc/model-zoo.git
Then upload a model to tracebloc:
from tracebloc_package import User
user = User() # log in with your tracebloc email + password
user.uploadModel("model-zoo/model_zoo/image_classification/pytorch/resnet_18.py")
Full walkthrough → run the training notebook in your own Jupyter
Index
model_zoo/index.v1.json lists every template with its task, framework, model_type, a display label, its default batch_size, a forecaster's look-back and horizon (forecast_window), its parameter count (PyTorch only) and whether it can be averaged across more than one data edge. It is generated by tools/build_index.py (make index) and checked in CI, so it always matches the templates.
Bring your own model
Not limited to this zoo. Any model that follows the model structure requirements works — PyTorch or custom containers.
Pretrained weights
Models in this zoo do not ship pretrained weights. Each model loads weights through its underlying library at runtime — torchvision's weights="DEFAULT", HuggingFace's from_pretrained(), or training from scratch.
If you're uploading your own model and want to bundle pretrained weights with it, name the weights file mymodel_weights.pkl and place it in the same directory as mymodel.py. The platform will pick it up automatically.
Links
Platform · Docs · Training notebook · PyPI package · Discord
License
Apache 2.0 — see LICENSE.
Questions? [email protected] or open an issue.
