Talks and Slides

This page collects external-facing KTransformers technical material. It is separate from task guides so users can read talks without confusing them with runnable installation instructions.

GOSIM 2026

The GOSIM talk covers workstation heterogeneous inference, CPU-GPU MoE resource mapping, expert deferral, layer-wise prefill, dynamic updates, and the local-inference-to-local-finetune direction.

When a talk contains an experimental feature, do not convert it into a user guide until the current repository entry and smoke result exist.