Cannot write files or execute commands using certain Local LLM
I am on Windows 11. I am running local LLM using llama.cpp. Using docker to run Hermes in sandbox.
I updated Hermes today, so client v0.21.5, commit 13dc3a7.
I did download qwen2.5-coder-14b-instruct LLM couple days ago. So it is relative new. I swore yesterday, I was able to write to /opt/data folder and execute codes. But today, it will only output snippets. It cannot even check the current working directory, or write a very simple test.txt file. It just provides snippet. It claims it did write the file, but there is no file. I ask if it can read the test.txt file, and it claims it can.
I tested 2 other LLMs (Qwen3.5-9B and gemma-4-12b), no issue on writing file and executing code.
Is there something I'm doing wrong with qwen2.5-coder-14b-instruct, or Hermes update broke something.
This is my llama.cpp call
"%SERVER_EXE%" ^ --model "%MODEL_PATH%" ^ --alias "Local Model" ^ --host %HOST% ^ --port %PORT% ^ --ctx-size 65536 ^ --cache-type-k q4_0 ^ --cache-type-v q4_0 ^ --device CUDA0,CUDA1 --split-mode layer --n-gpu-layers all ^ --main-gpu 0 ^ --flash-attn on ^ --no-mmap