Replies: 5 comments
-
Thanks for posting this, we are currently adding the pricing information in a patch release to make this available for everyone. There won't be a change if you already added the price manually. gpt4o uses a new tokenizer ( |
Beta Was this translation helpful? Give feedback.
-
js-tiktoken was just updated, I'll include token counts in this patch release already: https://github.com/dqbd/tiktoken/releases/tag/js-tiktoken%401.0.12 |
Beta Was this translation helpful? Give feedback.
-
@varenc released here by the way: https://github.com/langfuse/langfuse/releases/tag/v2.39.0 |
Beta Was this translation helpful? Give feedback.
-
Hey folks, also working on getting the new model into Playground and Evals asap :) Will report back here |
Beta Was this translation helpful? Give feedback.
-
Here we go: https://langfuse.com/changelog/2024-05-14-openai-gpt-4o-playground-evals |
Beta Was this translation helpful? Give feedback.
-
Describe the feature or potential improvement
OpenAI released their gpt-4o model today and it's now immediately available to all API users. It make sense for Langfuse to include this as one of the official pre-filled models. The docs say to request official support via an issue like this.
Additional information
There's both a
gpt-4o
and agpt-4o-2024-05-13
per the usual convention. Also I believe langfuse will need to update to the latest version 7.0 oftiktoken
which was just updated with gpt-4o support.Here's a screenshot of the working a model definition I create for myself:
But note: in the model's openai Tokenizer config I had to just claim the model was the older gpt-4-turbo. Setting
{ ..., "tokenizerModel":"gpt-4o-2024-05-13", ... }
resulted in gpt-4o logs showing 0 tokens, hence the need for atiktoken
upgraded.Beta Was this translation helpful? Give feedback.
All reactions