(Sorry if I ramble I've spent some time on token counting recently and have some opinions on it)
It's neat that it's client side only but it's quite feature lacking compared to https://tiktokenizer.vercel.app/ or even a small python script I wrote.
Feedback (Feature parity):
-Identification and separation of individual tokens. Useful for prompting/adjusting logit biases.
-Optional chat formatting
-It's not very responsive to type on, perhaps you could delay the token calc a little
Further suggestions (I don't know how simple you want to keep it):
-Counting tokens for Function Calling json functions
Yeah this is pretty much solved in every ecosystems but Javascript. It made me not use openAI's chat streaming because guessing became to hard. I'd rather show a loading and have precise measurements.
It's neat that it's client side only but it's quite feature lacking compared to https://tiktokenizer.vercel.app/ or even a small python script I wrote.
Feedback (Feature parity): -Identification and separation of individual tokens. Useful for prompting/adjusting logit biases.
-Optional chat formatting
-It's not very responsive to type on, perhaps you could delay the token calc a little
Further suggestions (I don't know how simple you want to keep it):
-Counting tokens for Function Calling json functions
helpful resources:
https://gist.github.com/CGamesPlay/dd4f108f27e2eec145eedf5c7...
https://hmarr.com/blog/counting-openai-tokens/