Dopp vs Hugging Face AutoTrain
Hugging Face AutoTrain trains a model from a dataset you upload, on their hardware, and puts the result on the Hub. Dopp captures the dataset from your live requests, labels it with the decision API you already use (Jev, for example), trains a small model that answers typed questions, measures it against that API, and routes requests between the two.
How is Dopp different?
No CSV. AutoTrain starts with "upload a dataset with a text column and a label column". Dopp's dataset is your traffic, recorded as it flows, in your app's exact request format, including many questions per request.
Labels come free. Jev answers every request; that answer is the label until you say otherwise. AutoTrain needs labels you already have.
Typed decisions, not just single labels. A request can carry several questions of different types (choices, yes/no, scores), including one per word of a text. Dopp trains and serves that shape as-is.
It closes the loop. AutoTrain ends with a model on the Hub. Dopp continues: a held-out number, live agreement on traffic, a routing switch with Jev as the fallback, and a browser copy.
Side by side
| AutoTrain | Dopp | |
|---|---|---|
| Dataset | CSV/JSONL you upload | Your live requests; plus paste, datasets, generated |
| Labels | You bring them | Jev, automatically; overridable |
| Task shape | One label per row (or standard tasks) | Typed questions per request, several at once |
| Training | Their hardware, per-hour | Our GPUs, price shown before the button |
| Result | A Hub model | A versioned model with a held-out number |
| Serving | Inference Endpoints (you set up) | Hosted per version; browser copy |
| Routing / fallback | None | Built in, with Jev |
| Who it's for | ML-adjacent teams with data | Product teams with traffic |
When AutoTrain is the better choice
- You have a labelled dataset and a standard task (sentiment, NER, image classification).
- You want the model on the Hub and full choice of base.
- You have someone to set up serving and evaluation.
When Dopp is the better choice
- You don't have labels, but you do have requests.
- Your questions are typed and several per request.
- You want to know when to switch, and a fallback when the model is unsure.
What AutoTrain does better
Breadth: image, tabular, LLM fine-tuning, many bases, and a model you can publish on the Hub. Dopp is narrow on purpose: typed decisions about a state.
Pricing
AutoTrain bills hardware time. Dopp shows the price of each step before you take it: labels, training minutes, serving. The browser copy serves for free.
Switch in three steps
- Route your decision traffic through Dopp.
- Train when you have a few hundred requests.
- Compare against Jev, then switch.
Questions
Can I upload my own labelled data too?
Yes; each row becomes a request with your label as the grounded answer.
Can I download the model to the Hub?
Yes, the checkpoint is yours to publish.
What base model is used?
Laya, a 0.4B encoder built for typed decisions, with more bases as needed.
What Dopp is. Dopp is a drop-in proxy for any Jev-compatible decision API: TypeSafe's Jev, Kev, GLiNER, simple-jev, or your own endpoint that answers typed questions about a state. You change one URL; every request still gets your current API's answer and is recorded as training data. When you have enough, one button trains a small open model on your own requests, measures it against the API it learned from on requests it never saw, and lets you choose who answers: the API, your model with the API as backup, or your model alone, on our servers, or offline, even in your users' browsers.