| copyright | lastupdated | subcollection | ||
|---|---|---|---|---|
|
2024-10-25 |
watsonx-code-assistant |
{{site.data.keyword.attribute-definition-list}}
Understanding {{site.data.keyword.cloud_notm}} subscriptions and resource units for {{site.data.keyword.wcaz_short}}
{: #wcaz-tokenz}
[{{site.data.keyword.wcaz_short}}]{: tag-dark-teal}
The {{site.data.keyword.wca_short}} cloud service works on a subscription basis. Learn about the relationship among your subscription, resource units, and tokens. {: shortdesc}
In generative AI, tokens are discrete pieces of data that, for language or code, can be a word, a character, or even a phrase (or sequence of words or characters). The models that underlie generative AI solutions predict content based on an understanding of context (which are previous tokens). When you query a model, it uses a number of tokens to accurately generate a response to your query.
When you use the {{site.data.keyword.wcaz_full_notm}} catalog page{: external}, what you're purchasing is entitlement to a number of resource units in your {{site.data.keyword.cloud_notm}} subscription. Each resource unit is equal to 150,000 tokens. For {{site.data.keyword.wcaz_short}}, COBOL applications require approximately 20-30 tokens per line of code. That value might vary depending on the application. Your tokens yield approximately 5000-7000 lines of code. That number can vary because the number of tokens that are required to generate a good response varies. When you run out of tokens, you can renew your contract to refill your credits.
Your {{site.data.keyword.cloud_notm}} subscription credits don't expire and you can use them on any service instead of, or in addition to, {{site.data.keyword.wcaz_short}}. Resource units are tied to usage, not a time frame. When you run out, you can resubscribe by renewing your contract. {: important}