-
Notifications
You must be signed in to change notification settings - Fork 13.3k
Closed
Labels
Description
Name and Version
% ./build/bin/llama-server --version
version: 6138 (60a76588)
built with cc (GCC) 15.1.1 20250729 for x86_64-pc-linux-gnu
I use gpt-oss with --jinja
and I tried using --reasoning-budget 0
, but it didn't change anything. I assume it is because the --reasoning-format
does not exist for it yet, but I didn't spend enough time at looking how the code works to know.
Operating systems
Linux
GGML backends
BLAS
Hardware
i5-8250U
Models
https://huggingface.co/unsloth/gpt-oss-20b-GGUF/blob/main/gpt-oss-20b-UD-Q6_K_XL.gguf
Problem description & steps to reproduce
llama-server --jinia --model gpt.gguf --reasoning-budget 0
Does nothing. It still "reasons"
First Bad Commit
No response
Relevant log output
llama-server --port 50050 --host 0.0.0.0 --model /mnt/nvme/models/gpt/gpt-oss-20b-UD-Q6_K_XL.gguf -c 0 --jinja --reasoning-budget 0