GitLab

gpt-oss works best with a context length of at least 8k. However,
for GPUs with limited amount of VRAM, there is a significant
performance hit to this increased context. In these cases, we
switch to the Ollama default of 4k