Unexpected empty grammar stack after accepting piece: G

[starbuckscoffee]

Which version of LM Studio?
LM Studio 0.3.23 (Build 3)

Which operating system?
Windows 11

What is the bug?
LLM: openai/gpt-oss-120b

After loading GPT-OSS-120B,
GPU Offload 36/36
Context Length 8192
Offload KV Cache to GPU Memory : OFF

The model outputs my 1st input correctly as expected.
When I typed 2nd query, the model outputs GGGGGGGGG forever...

After I cleared all messages, I typed 2nd query again, then the model outputs correctly as expected.

Screenshots

Logs
Unexpected empty grammar stack after accepting piece: G

To Reproduce
Described each steps in "What is the bug" section.

[waltercool]

Anyone knows why this happens?

It seems to happen randomly, and after certain random threshold it just repeats letter G no matter what.

GPT Offload 36/36
Context Length: 30733
Offload KV Cache to GPU Memory: ON
K Cache Quant Type : F16
GGUF: Vulkan llama.cpp (Linux) v1.48.0
Harmony: v0.3.4

This is regardless of the Reasoning Effort

[starbuckscoffee]

I close it because it does not relating to LM Studio.

The issue was discussed in a thread of llama.cpp: ggml-org/llama.cpp#15274
Then the issue was passed to PR15619 : ggml-org/llama.cpp#15619
and PR15652 : ggml-org/llama.cpp#15652

Please monitor these threads above...