bug: LlaVa 7B responds as though I've provided JSON data

[erayerdin]

Jan version

0.5.3

Describe the Bug

I've downloaded LlaVa 7B and, to test it, I've sent a simple hello message, and it responded with:

The user provided a JSON object containing a "name" field and a "type" field. The "name" field is a string, while the "type" field is a string as well. The "name" field contains the word "Rock". The "type" field contains the word "type".

Based on the provided JSON object, the "name" field contains a single word and the "type" field also contains a single word. The "name" field is a string, and the "type" field is a string. The JSON object is a valid representation of two pieces of data, a name and a type, but without additional context, it is difficult to determine the specific use case for this data.

Steps to Reproduce

1. Download LlaVa 7B
2. Write a message.
3. Get the weird response.

Screenshots / Logs

What is your OS?

• MacOS
• Windows
• Linux

[imtuyethan]

Tested on:

Linux:

107 (test-ubuntu-app-cpu-2)

Mac:

Operating System: MacOS Sonoma 14.2
Processor: Apple M2
RAM: 16GB

Results:

On Linux:

On Mac: Its responses vary & not stable:

High chances are this could be because of model performance, or our model settings are outdated? cc @louis-jan

[imtuyethan]

Related: janhq/models#46

[imtuyethan]

I'll close this issue and follow up on janhq/models#46 instead