​

Start Here ↓

​

Create an New Evaluation

To create a new evaluation, you need to have a fine-tuned model. If you don’t have a fine-tuned model, you can create one by following the steps in the Fine-Tuning section.

You need to fill out the following fields:

• Evaluation Prompt: The prompt to use for the evaluation.

​

Example of an Evaluation Prompt

Convert medical reports into ICD-10 codes.

You will be given a textual medical report. The model must return a list of ICD-10 codes.

Evaluate:

Are the codes valid and from the ICD-10 standard? Do the codes align with the diagnosis mentioned in the report? Are there unnecessary or incorrect codes predicted?

• Dataset: The dataset to use for the evaluation.
• Snapshot: The snapshot of the dataset to use for the evaluation.
• Models: You can select multiple models to evaluate.

Once you have filled out the fields click the “Create Evaluation” button.

​

Check Your Results

Once you have created an evaluation, you can check your results by clicking on the evaluation and each datapoint to get more details.

​

Here are Some Results to Keep in Mind:

• Average Score: The average score of the evaluation for the model.
• Model Name: The name of the model the evaluation was run on.
• System Prompt: The system prompt used for the evaluation.
• User Message: The user message from the dataset.
• Original Assistant Message: The original assistant message from the dataset.
• Predicted Assistant Message: The predicted assistant message from the model.
• Model Score: The score of the model chosen for the evaluation.
• Score Reason: The reasoning behind the score.

​

Next Step: Fine-Tuned Model → Playground