ByteDance
Posted 1mo ago

推荐大模型训练/推理工程师 - Data-抖音/直播/电商/剪映

ByteDance
Beijing or Shanghai or Hangzhou
OnsiteFull Time
Responsibilities
  • developing models
  • optimizing inference
  • designing architecture
Requirements
  • Bachelor's (2027 graduates) in CS or software engineering
  • Strong programming in C/C++/Python/Java
  • CUDA experience
  • Familiarity with PyTorch/TensorFlow
  • TensorRT/Triton/CUTLASS
  • LLM training/inference tools (Megatron
  • Deepspeed, vLLM
  • TRT-LLM), and strong troubleshooting skills
Technical tools mentioned
PyTorchTensorFlowCC++PythonJavaCUDATensorRTTritonCUTLASSMegatronDeepspeedvLLMSGLangTRT-LLM

Job description

团队介绍:推荐架构团队支撑字节跳动旗下多款APP产品,如抖音、今日头条、番茄小说、西瓜视频、剪映等推荐系统架构的设计和开发,保障系统的稳定和高可用,致力于抽象系统通用组件和服务,建设推荐中台、数据中台;关于在线服务,在这里你有机会参与大规模机器学习在线预估框架的研发与优化,也有机会参与模型训练与调度等相关问题的研究与突破,解决系统瓶颈,降低成本开销;如你对大数据感兴趣,在这里也有机会参与通用实时计算系统的开发、构建统一的推荐特征中台,为推荐业务实现先进的消重、计数、特征服务等;我们期待热爱技术的你加入,一起创造更多可能。

1、完成LLM训练、推理技术在推荐大模型场景的研发、落地和调优,解决推荐场景的大模型工程问题;
2、针对PyTorch、TensorFlow等框架提供高自动化、极致性能的模型优化方案;
3、推动基于大模型开源组件构建,设计和实现新一代推荐系统架构,支持推荐大模型在业务落地;
4、与算法团队Co-Design,推动端到端生成式推荐技术创新的预研和落地。

1、2027届获得本科及以上学历,计算机、软件工程等相关专业优先;
2、扎实的编程和算法基础,熟练掌握C/C++、Python、Java等编程语言,有CUDA开发经验,熟悉TensorRT/Triton/CUTLASS经验者优先;
3、熟悉大模型推理加速的研究和技术进展,包括但不限于模型量化、图编译、算子调优、KVCache、PD分离等;
4、掌握LLM训练/推理技术,有实际开发和应用落地经验,例如Megatron、Deepspeed、vLLM、SGLang、TRT-LLM等;
5、对大模型工程有丰富的经验和广阔视野,紧追开源动态,具备独立的Trouble-shooting能力。

About ByteDance

Developing AI-driven content platforms and mobile applications.

Similar jobs

Large Model Training and Inference Engineer roles near Beijing, Beijing
1mo
Save
Mark Applied
Hide
大模型训练推理研发工程师—大装置
Beijing or Shanghai
OnsiteFull Time
SenseTime
SenseTimeHKEX: 0020: Develops artificial intelligence software and computer vision technology.
Bachelor's in CS or related, strong CS fundamentals, proficiency in Pytorch and C++, familiarity with large-model frameworks and performance optimization, strong communication and problem-solving.
Pytorch, C++, sglang, vllm, tensorrt, megatron, cuda
2mo
Save
Mark Applied
Hide
顶尖应届-具身大模型训练推理工程师-机器人事业部
Beijing, Beijing, China
OnsiteFull Time
Xiaomi
XiaomiHong Kong Stock Exchange: 1810: Designs and manufactures smartphones, consumer electronics, and electric vehicles.
Design and implement distributed training/inference infrastructure for large embodied models; strong Python/C++ and Linux skills; deep knowledge of PyTorch distributed frameworks and GPU/CPU/memory/network optimization.
PyTorch, FSDP, DeepSpeed, Megatron, Megatron-LM, BF16, FP8, FlashAttention, Triton, Python, C++, Linux, NCCL, RDMA, InfiniBand, CUDA, TensorRT, ONNX Runtime