chat stuck in infinite loop

[sam-coffey]

Have tried installing and reinstalling many times, chat with any agent always gets stuck in loop where agent just keeps replying to itself.

Happens each time I successfully start an agent with npm start (either chatting in terminal in older versions or with localhost:5173)

I would expect the agent to have a dialogue with me but it becomes impossible when the agent just keeps saying things to itself over and over.

[madjin]

This happened to me, result of an outdated characterfile. Try a new one?

[sam-coffey]

Thanks - I have isolated my issue to the llama local model. I tried with a gemini API key and everything worked as expected. If I set a character to use the Llama local model then that is when I get into the infinite loop type issue.

[sam-coffey]

I tried editing one of the characters in the characters folder to use "llama_local" as the modelProvider but that resulted in the same behavior of the model just talking to itself rather than waiting for further input from the user.

[kmozurkewich]

same issue here on local machine and h100 with 22.04.

:// Show less responses...
```json
```json
:// Add your response here
```json
```json
```json
```json
```json
```json
```json
```json
```json
```json`

[brogovia]

Same issue here using a local model with cuda.

[mosayeri]

same issue with an ollama local

[BrandonFlorian]

I am having the same problem trying to run locally.

[tercel]

same issue with llama_local

[zoe27]

meet the same issue when running locally with llama_local model

[hiteshjoshi1]

Filed #1575 , seems like the same issue. Will close that in favor of this one.

[briancullinan2]

not using llama_local but still seeing this error, tried gpt-4o and meta-llama and hermes, all same result, repeats endlessly expecting a different response from the model like it isn't formatting properly.

EDIT: forgot to specify, using Apple M1.

EDIT: i left out the --character="..." part and got the error, now i added it back in and get nothing. no server request detected.

EDIT: i removed --character again, and changes to gpt-4o-mini and finally got a single response # Response { "user": "Eliza", "text": ".... Is there a bug in loading characters maybe?

EDIT: here's a gist of what i'm dealing with: https://gist.github.com/briancullinan2/fed55ab2f982baebda5f32e1095e7b3f

[AIFlowML]

Meanhile llama_local is known to have issues genrally other models not.
Can you please try with the latest version and report back ?

[danielNg25]

Any update? I'm facing the same issue

[zoe27]

seems it’s caused by the response generation logic in llama.ts. I fixed it locally, it seems to work, and submitted a PR

[modddypan]

seems it’s caused by the response generation logic in llama.ts. I fixed it locally, it seems to work, and submitted a PR

it works for me with v0.1.7-alpha.2

[BrandonFlorian]

seems it’s caused by the response generation logic in llama.ts. I fixed it locally, it seems to work, and submitted a PR

I just tried your PR and it works to fix the infinite loop and get the chat messages to the front end :)

I am still getting the exact same responses every time from the model no matter what question I ask, but that may be a different issue.

[zoe27]

I am still getting the exact same responses every time from the model no matter what question I ask, but that may be a different issue

seems things goes well in my side

[BrandonFlorian]

I am still getting the exact same responses every time from the model no matter what question I ask, but that may be a different issue

seems things goes well in my side

Thank you for confirming and for fixing the bug.

It seems I have a separate issue to figure out.

[AIFlowML]

Fixed bu @zoe27 with PR
User confirm it work.
Closing this issue now.