-
|
Does koboldcpp have a detokenize API endpoint or any way to split an input string up into its plaintext tokens? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 5 replies
-
|
API is described in built-in page: http://localhost:5001/api |
Beta Was this translation helpful? Give feedback.
-
|
What I'm looking for is an option to turn the token ids back into text. But it doesn't look like it has an endpoint for that currently. |
Beta Was this translation helpful? Give feedback.
-
|
A detokenization endpoint does not currently exist. What's the use case for it? |
Beta Was this translation helpful? Give feedback.
-
|
I'm implementing token and phrase biasing for mikupad, and I wanted to display the biased strings split into their tokens. |
Beta Was this translation helpful? Give feedback.
-
|
I see. I'll consider it for the future - it's not so straightforward due to multibyte token sequences. |
Beta Was this translation helpful? Give feedback.
-
|
Got it. Thanks for letting me know. |
Beta Was this translation helpful? Give feedback.
API is described in built-in page: http://localhost:5001/api
/extra/tokencount - counts number of tokens and (in recent enough version) sends tokens ids of input text.