You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The token.js package currently maintains a complete list of all models available from each supported provider. This provides the following benefits:
Full type support when specifying model names
Improved UX when a given model does not support a specific feature. For example, if a model does not support JSON output and the user specifies JSON output, then token.js will throw a more consistent error earlier in the process instead of waiting for the underlying API to error out.
However, it also has the following drawbacks:
New model releases are not immediately supported. An update to token.js is required to enable any new model.
Non-public models are not supported (fine tuned models mainly)
Solution
We should provide some method for bypassing the model name checks, so that users are able to use new models immediately upon release. token.js should make a best effort to continue to maintain a comprehensive list of models and their supported features to provide good UX, but it should not prevent the user from accessing models that are not explicitly supported.
Community Feedback
I'm opening this ticket to solicit community feedback on the best UX for resolving this problem. Currently, we have a PR open from @shadokan87#127 which resolves this issue by allowing the user to call a new function tokenjs.extendModelList to extend the list of supported models. It also allows the user to define supported features for their self defined models which seems like a great feature to me. After some consideration, this implementation seems pretty solid to me but I'm hoping we can get some input from other token.js users before moving forward with it. If you have thoughts on how to resolve this problem, please leave a comment on this issue.
The text was updated successfully, but these errors were encountered:
Problem
The
token.js
package currently maintains a complete list of all models available from each supported provider. This provides the following benefits:token.js
will throw a more consistent error earlier in the process instead of waiting for the underlying API to error out.However, it also has the following drawbacks:
token.js
is required to enable any new model.Solution
We should provide some method for bypassing the model name checks, so that users are able to use new models immediately upon release.
token.js
should make a best effort to continue to maintain a comprehensive list of models and their supported features to provide good UX, but it should not prevent the user from accessing models that are not explicitly supported.Community Feedback
I'm opening this ticket to solicit community feedback on the best UX for resolving this problem. Currently, we have a PR open from @shadokan87 #127 which resolves this issue by allowing the user to call a new function
tokenjs.extendModelList
to extend the list of supported models. It also allows the user to define supported features for their self defined models which seems like a great feature to me. After some consideration, this implementation seems pretty solid to me but I'm hoping we can get some input from othertoken.js
users before moving forward with it. If you have thoughts on how to resolve this problem, please leave a comment on this issue.The text was updated successfully, but these errors were encountered: