1 / 5
You take a pretrained model and adapt it to your specific data with a small extra training run. You ___ the model.
-
-
-
-
Fine-tune a model means further training a pretrained model on task-specific data.
- train a model — from scratch or initial training
- fine-tune a model — adapt an existing one
"Sharpen off" and "refit out" aren't the collocation. Example:
"We fine-tuned the base LLM on our support tickets."
2 / 5
The trained model is packaged and put into production behind an API. The team ___ the model.
-
-
-
-
Deploy a model means putting it into production so it can make predictions.
- deploy a model — release it to prod
- Then you serve it — handle inference requests
"Plants up" and "sets out" aren't standard. Example:
"We deployed the model behind a REST endpoint."
3 / 5
In production, the system handles incoming inference requests from the model. It ___ the model.
-
-
-
-
Serve a model means hosting it to respond to inference/prediction requests.
- serve a model — handle live predictions
- Done by a model server (e.g., TF Serving, Triton)
"Feeds off" and "runs up" aren't the collocation. Example:
"We serve the model with autoscaling GPU pods."
4 / 5
Over time, real-world data changes and the model's accuracy quietly drops. The model is said to ___.
-
-
-
-
Drift describes a model's performance degrading as production data diverges from training data.
- model drift / data drift / the model drifts
- Monitored so you know when to retrain
"Slip off" and "sag out" aren't standard. Example:
"Accuracy fell as inputs drifted from the training distribution."
5 / 5
Because the model has drifted, the team trains it again on fresh data. They ___ the model.
-
-
-
-
Retrain a model means training it again, usually on newer data, to restore performance.
- retrain a model — periodically or when drift is detected
- Often automated in an MLOps pipeline
"Re-teach up" and "re-fit out" aren't standard. Example:
"We retrain the model weekly on the latest data."