具身世界模型推理INFRA工程师 - XiaomiRobotics

- Adapt multi token prediction for inference - Apply FP8 quantization for models - Apply NVF4 quantization for models - Collaborate with algorithm team to optimize production rollout and open source release - Implement tensor parallelism and expert parallelism for inference framework - Integrate CFG parallelism for inference - Optimize inference throughput for embodied world models