gptscript-ai / desktop

MIT License
22 stars 13 forks source link

UI - Seeing erroneous response from Anthropic provider once we confirm the calls to provider repo. #121

Closed sangee2004 closed 2 months ago

sangee2004 commented 4 months ago

gptscript version v0.0.0-dev-53f7fbde-dirty

Steps to reproduce the problem:

  1. Launch UI with default model set to Anthropic provider - gptscript --ui --default-model 'claude-3-5-sonnet-20240620 from github.com/gptscript-ai/claude3-anthropic-provider' github.com/gptscript-ai/cli-demo
  2. Once UI is launched and prompted for allowing call to Anthropic provider repo , allow this operation.
  3. Notice that there is the following chat response message after allowing this.
Screenshot 2024-06-26 at 11 57 52 AM
tylerslaton commented 2 months ago

This is old and likely just an LLM issue. Can you try replicating this now?

sangee2004 commented 2 months ago

This issue is still seen when testing with electron app commit id - c68bec95

Screenshot 2024-08-07 at 12 44 30 PM

The assistant used :

Name: sangeeclaudeprovider
Type: Tool
Model: claude-3-5-sonnet-20240620 from github.com/gptscript-ai/claude3-anthropic-provider
Chat: true

You are a good assistant.
tylerslaton commented 2 months ago

I think that this is similar to the issue we saw where the gateway provider output its address @cjellick . However the user cannot currently enter this state since anthropic isn't on our current model list.

sangee2004 commented 2 months ago

Tested with electron Version - 0.10.0-rc3

I dont see this issue any more.

  1. Create an assistant with model (in Advanced) as claude-3-5-sonnet-20240620 from github.com/gptscript-ai/claude3-anthropic-provider
  2. Start a chat using this assistant.
  3. I am prompted to enter anthropic key and I am able to get response successfully.

There is no erroneous response from Anthropic provider seen when confirming calls to the provider repo anymore.