Latest vulkan version doesn't follow instruction

[duynt575]

What happened?

I have a prompt file with content like this:

You are a friendly assistant. USER: Hi ASSISTANT: Hello.</s> USER: Who are you? ASSISTANT: I am WizardLM.</s>

I use -f to load it and it works with both cuda and vulkan version (WizardLM-2-7B-Q8_0). But latest vulkan version seems to break this. The model keeps asking what I want to ask even though I ask it repeatedly to do something, it's like it can't see what I am asking. Reverting back to b3150 and it works again without any issue.

Name and Version

version: 3155 (c8a8219)

What operating system are you seeing the problem on?

Windows

Relevant log output

No response

[github-actions]

This issue was closed because it has been inactive for 14 days since being marked as stale.