[lxe]

Seems to be significantly slower than on the official deepseek API. But at least it's the full model!