Fix model switcher for Azure #508
Conversation
Updated ChatInput.tsx text area to autoextend for larger prompt size
Correct the improper context being used when it exceeds the limit.
User message should not render as Markdown
Co-authored-by: dev <dev@ptfprod7>
* feat: Add conditional search input for conversation list * style and ux adjust --------- Co-authored-by: Tekle Ayele <[email protected]>
* Update .gitignore * Update next.config.js * Enable GPT-4 * Update index.ts * Fix logic * Fix error * Add set openai key instruction * Fix menu safari * Revert changes * new line --------- Co-authored-by: Sonny <[email protected]>
* mobile ui updates * fixes sidebar btn * return if null * mobile input blur * handle mobile enter key * new convo name * new delete mechanism * test height * revert * change padding * remove overflow * check relative * padding * done * retry * test * test * should work now * test * test * more * max h * revert * done
* use tiktoken for api limit * model conditional char limits on frontend * adjust for completion tokens --------- Co-authored-by: Alan Pogrebinschi <[email protected]>
* change search language * add to import/export * fix log
- conserve space in new convo view - fix overlapping edit and delete chat message icons - fix overlapping copy icon
basic secuirty policy as the repo didn't have one yet, update as needed.
|
This isn't correct. The deployment name in the URL is unrelated to the model name and must be configured. I don't think there is a way to get a list of available deployments unless you can access the Azure API. |
|
@kleptog it is correct. The azure api returns both the model name and the deployment name. The code already checks against the model (model field), then translates it to the deployment name (id field) for use in the URL. See https://github.com/mckaywrigley/chatbot-ui/blob/main/pages/api/models.ts#L53. With this PR it doesn't matter what the deployment is called, as it looks for the model and updates the URL accordingly. |
|
@mckaywrigley I’ve fixed merge conflicts. Would it be possible to get this reviewed and merged? |
|
This works! Great job @itbm |
|
It looks like the Azure deployment name would need to match the model name for this to work. Is that correct? I'm not sure why someone would use a deployment name that doesn't match the model name in setting up Azure OpenAI, but it is possible. Perhaps this is useful for fine tuned models, I'm not sure. |
|
With this PR they can be different. Chatbot UI will use the model name internally up until it makes the request to Azure, at which point it will use the deployment name in the URL. |
|
Currently always the default model (configured in |
Still facing this issue, when deployment name is not = model name the default model assigned at start of conversation is not assigning the model id from azure instead gives fallback one. if one of the deployments is named gpt-35-turbo it is able to work fine for other deployments which dont have name same as model name |
Reverts #499
I suggest reverting this as it isn't necessary and breaks model switching between GPT-3.5, GPT-4 and GPT-4-32K in the UI. Should be merged with #509.
See #498 (comment)