Rules
Last updated
Last updated
There are certain rules to consider when designing chains.
A token is a unit of text processable by language models. You can think of tokens as pieces of words each around 4 English characters in length. Language models have limits on how many tokens they are able to process and generate on each run. The limits are as follows:
GPT-4: 8192
GPT-3.5: 4096
A valid chain is one without any cycles along the path of any set of connected flow edges. A cycle, if allowed, would cause an infinite execution loop. FireKit detects cycles at runtime and will raise an error if detected.
Skill chains follow the rules of regular chains with these exceptions:
A flow edge from a block outside a skill to a block within a skill is a valid connection. In other words, the output of a block outside of a skill can be used as input for a block inside a skill
The output of a block inside a skill cannot be used as input for a block outside a skill
With regards to cycle detection, all blocks within a skill, along with the agent block, can be viewed as a single entity.
