A test credit can be used to run tests on models in KraspAI. To run a test, the test's prompt is given to the model and it's response is recorded. Depending on the length of the prompt and the model's completion, a different amount of test credits is used. See the next question for more details.

Whether we consider a test short, medium or long is determined by the total number of tokens in the prompt and the model response. A test is considered 'short' if this sum is less than 100. A test is considered 'medium' if this sum is between 100 and 500. A test is considered 'long' if this sum of is greater than 500.

The number of test credits needed to run a test depends on the length of the test. A short test costs 1 test credit, a medium test costs 5 test credits and a long test costs 10 test credits.

You can use any model that is available in KraspAI. The currently available models are:

openai-gpt-3.5-turbo, openai-gpt-4, openai-gpt-4-0314, openai-gpt-3.5-turbo-0301, openai-text-davinci-003, openai-text-davinci-002, openai-text-curie-001, openai-text-babbage-001, openai-text-ada-001, cohere-command, cohere-command-light, alephalpha-luminous-base, alephalpha-luminous-extended, alephalpha-luminous-supreme, alephalpha-luminous-base-control, alephalpha-luminous-extended-control, alephalpha-luminous-supreme-control, meta-llama-2-70b-chat, tii-falcon-40b-instruct

These are the currently available models. Over time availability of models may vary, as we depend on access to the models' APIs – which may be shut down or inaccessible for us for reasons beyond our control. Reach out to support@krasp.ai to request a model we don't support yet.