Omni模型算法工程师/研究员

Xpeng

北京、深圳

≈ 481,1 тыс.–481,1 тыс. ₽ · 6,1 тыс.–6,1 тыс. $ · 5,4 тыс.–5,4 тыс. €

41 667 – 41 667 CNY

🏢 Офис

Middle

Полная занятость

Продуктовая

Китай

Tech

Описание вакансии

Обязанности

•Apply reinforcement learning for training

•Build end to end audio to audio models

•Construct multimodal data pipelines

•Convert instructions into action sequences

•Create speech to action models

•Define data quality and sampling strategies

•Deploy models to edge and heterogeneous compute

•Design multimodal model architectures

•Develop embodied agent and task planning systems

•Develop multimodal alignment techniques

•Evaluate model ability safety latency and resource use

•Implement temporal modeling and long context understanding

•Integrate models with robot perception cognition decision and control

•Optimize models with quantization distillation pruning and operator optimization

•Perform instruction tuning and preference alignment

•Train using pretraining and finetuning

•Unify representations across text speech vision video action

Технологии: Agent planning, Audio Processing, Audio to Audio, C++, Context understanding, Data Engineering, Distributed Training, Edge Computing, Embodied AI, End to End, End-to-end learning, Heterogeneous computing, Human Feedback, Instruction Tuning, Knowledge Distillation, Language Models, Large Language Models, Learning from Human Feedback, Linux, Long Context, Long Context Understanding, Model Compression, Model Deployment, Model Evaluation, Model Training, Multimodal Learning, Operator optimization, Preference optimization, Pretraining, Pruning, PyTorch, Python, Quantization, Reinforcement Learning, Reinforcement Learning from Human Feedback, Speech Recognition, Speech to Action, Task planning, Temporal modeling, Transformer, Vision Language Models, Vision-language