SenseTime
Posted 2w ago

训练推理优化实习生,多模态

SenseTime
Beijing, Beijing, China
OnsiteInternship
Responsibilities
  • developing modules
  • optimizing models
  • analyzing performance
Requirements
  • Master's or PhD students in CS/AI/Math proficient in C++ and Python
  • Solid data structures and algorithms
  • Experience with AI coding assistants (Codex
  • Claude Code
  • Cursor)
  • Familiarity with model training/inference and performance analysis
Technical tools mentioned
C++PythonCodexClaude CodeCursorsglangvllmStable DiffusionComfyUIMidjourneyLightX2VRDMAIBRoCE

Job description

1. AIGC研究与工程探索
· 深度参与商汤开源lightx2v引擎(https://github.com/ModelTC/LightX2V/)的社区核心模块开发与调优。
· 跟踪视频/图像生成及LLM推理的前沿技术趋势,将学术界最新成果转化为引擎的实际性能增益。
· 协助头部客户快速完成业务落地,形成规模化应用效应。
2. 模型性能剖析与国产化深耕
· 熟练分析训练推理过程中的性能瓶颈。
· 基于国产算力(沐曦、寒武纪等)生态,开展算子级深度优化与工具链整合,解决精度对齐、算子性能不足等国产化迁移痛点,为客户提供平滑迁移的端到端保障。

1. 计算机、人工智能、数学等相关专业在读硕士、博士。
2. 熟练掌握 C++ 和 Python 语言。
3. 有扎实的计算机基础知识,熟练掌握常见的数据结构和算法。
4. 具备 AI 辅助编程实践经验,能够熟练使用 Codex、Claude Code、Cursor等 IDE 编程助手,并能够将其应用于实际研发场景提升开发效率。

加分项
1. 对AI大模型相关核心技术原理有认知,熟悉sglang、vllm等大模型框架,有训推性能优化经验。
2. 熟悉视频和图像生成(x2v),熟练运用Stable Diffusion、ComfyUI、Midjourney等工具链,掌握lightx2v等推理框架。
3. 熟悉国产算力(沐曦、寒武纪等)生态与软件栈,具备算子适配、性能调优及精度对齐的实操经验。
4. 有高性能网络 (rdma/ib/roce) 相关的经验。

About SenseTime

Develops artificial intelligence software and computer vision technology.

Similar jobs

Training and Inference Optimization Intern roles near Beijing, Beijing
1mo
Save
Mark Applied
Hide
训练推理优化实习生(Mstar)
Beijing or Shanghai or Suzhou
OnsiteInternship
Momenta
MomentaHong Kong Stock Exchange: 06880: Develops autonomous driving software and intelligent transportation solutions.
Master's degree in computer-related field, strong C++/Python/CUDA skills, experience with distributed training and communication libraries (NCCL/RDMA), and familiarity with large model training and low-precision strategies.
CUDA, NCCL, RDMA, Python, C++, CUTLASS, Triton, PyTorch, Megatron-LM, DeepSpeed, VeRL, SGLang, vLLM
7mo
Save
Mark Applied
Hide
推荐大模型训练/推理优化实习生-Data-抖音/直播/电商/剪映
Beijing, Beijing, China
OnsiteIntern
ByteDance
ByteDance: Developing AI-driven content platforms and mobile applications.
Seeking 2027 undergraduate+ students with strong programming, algorithms, LLM training/inference and CUDA experience; familiarity with PyTorch/TensorFlow and model optimization techniques.
PyTorch, TensorFlow, C/C++, Python, Java, CUDA, TensorRT, Triton, Cutlass, Megatron, DeepSpeed, vLLM, TRT-LLM, SGLang