c0demunk33 / bespoke_automata Goto Github PK

Can we make the list of terms in the random text node configurable by file.

Companion to: #25 (comment)

Proposal to implement porting of old brain schema's to the current node referencing schema.

Old schemas should be detected on loading, ported and the user notified to save in the new format.
Node reference mapping should be driven from a configuration file.

See companion issue for mapping table.

Could we add a switch to the array and dictionary assembler to only add an item if it's unique?

The use case I have in mind is merging list outputs from 2 different GPT Text Nodes.

Thanks.

To complement arrays and dictionaries, can we have a stack with pusher and popper

I'd like to propose a node that enables a user to include ML models in a Brain configuration.

As a starter for 10, the node could work as follows:

• The node executes a local script in an appropriate folder
• The node passes a JSON object as an input
• The node accepts a JSON object as a return
• The user is responsible for ensuring their environment is appropriately configured
• The node should be configurable to set an appropriate timeout

Discussion and feedback welcome

This one affects the LLM node with a buffer length of 0.

When running a brain in Bespoke Manager, the LLM call is incomplete when the brain is invoked the first time. On the second pass the buffer has reset to 10.

Brain then fails when you run out of context length.

Proposal to rationalise the node categories and improve node labelling. I suggest the following categories:

• Instruction
• Action
• Generation
• IO
• Utility

See attached mapping table for as-is and to-be node reference labels and node titles. Feedback welcome.

This change will break compatibility with brain schema's developed up till now. There is a companion issue raised to automatically port old brain schema's to the new node references. #26 (comment)

Also propose to change the label of Group to Cortex

RefsAndTitles.csv
RefsAndTitles2.csv