Frequently Asked Questions

​

Privacy & security

​

Can the data scientist see my raw data?

No. Data scientists submit model code; the code runs in your secure environment, against your data, in an isolated sandbox. They see the evaluation metrics and, if Weight Files is on, the trained model weights (see below). Raw data never leaves your infrastructure.

​

What gets shared with the data scientist or with tracebloc?

Raw data never leaves your infrastructure. What goes to tracebloc is the evaluation metrics (accuracy, F1, and so on) and the trained model weights: after each training cycle, your secure environment sends the model’s weight updates to tracebloc, which combines them across secure environments (federated averaging).

​

Where do trained model weights end up?

Your secure environment trains the model locally and sends the weight updates to tracebloc after each cycle, where they are combined. Data scientists can download the trained weights of their own and their team’s experiments unless you turn off Weight Files (Allow participants to download weight files) in the use case’s admin settings. It is on by default.

​

Is the secure environment’s egress restricted?

Yes — the chart applies a Kubernetes NetworkPolicy that only allows training pods to reach the tracebloc backend (for orchestration and the per-cycle weight updates) and the in-cluster proxy that handles result and FLOPs reporting. The egress lockdown only takes effect on a CNI that enforces NetworkPolicy. EKS’s default VPC CNI does not, out of the box — see the EKS deployment guide for what to install.

​

Infrastructure

​

Do I need a GPU?

For most ML workloads, yes. Training runs on NVIDIA GPUs of the Turing generation or newer; AMD, Intel and Apple GPUs train on CPU. On Linux, the standalone installer detects an NVIDIA GPU and installs the driver automatically. See GPU support for supported GPUs, drivers and platforms. CPU-only is fine for small tabular and text models, but expect long training times on anything image- or sequence-heavy. Image tasks do train on a CPU-only machine, slowly; don’t ingest a dataset while a training runs there, as both compete for the same CPU.

​

What Kubernetes versions do you support?

Kubernetes 1.24 and above. See the setup guide for the full prerequisites.

​

Can I run on-premise only?

Yes. The standalone installer provisions a local cluster on your machine, or you can deploy the unified Helm chart into your existing on-prem Kubernetes. Your secure environment never depends on cloud infrastructure for training.

​

Can I install offline or run behind an egress proxy?

You can install from a packaged chart .tgz rather than the public chart repo, and platform traffic can route through your corporate proxy (set HTTP_PROXY / HTTPS_PROXY before installing). The environment is outbound-only — nothing needs to reach in — but it does need outbound HTTPS to the tracebloc platform to run experiments, so it is not fully air-gapped. What you get instead is defined, auditable ingress and egress: raw data never leaves your infrastructure; only metrics and model weight updates are sent out. See the install guide.

​

Training & models

​

How do I monitor training?

Through the tracebloc dashboard — every experiment, every model, every metric.

​

What if an experiment fails?

Your secure environment retries failed requests to tracebloc automatically. If the experiment still fails, the dashboard marks it Failed, with the error category (for example data, model, or out of memory) and the reason. For ingestion-time failures, check the Job logs in the namespace you deployed into.

​

Can I bring my own model?

Yes. Use the tracebloc Python package to upload a model file. Supported frameworks are PyTorch, scikit-learn (including XGBoost, CatBoost and LightGBM), lifelines, and scikit-survival; there is no container upload. For ready-made starting points, see the model zoo. TensorFlow is deprecated — new TensorFlow uploads are no longer accepted, though existing TensorFlow experiments remain readable.

​

Do you support fine-tuning?

Yes — the same upload flow handles full training, fine-tuning with pretrained weights, and inference-only evaluation.

​

Cost & support

​

How much does it cost?

See the pricing page on the website.

​

How do I get help?

• Email [email protected]
• Open an issue on the relevant GitHub repo: client, docs
• Join the Discord

​

Next steps

• Get started — install your secure environment
• Browse key terms
• Use the Python SDK