Responsibilities
- 训练平台搭建：负责作业调度、断点续训、监控告警及数据/权重流水线建设
- 优化数百卡规模的分布式训练性能（如 FSDP、张量并行、Context Parallel、Ulysses）
- 构建 PB 级视频数据 pipeline，处理 NVDEC 解码、VAE latent 缓存等
- 通过 FlashAttention、FP8 混合精度、Triton kernel 等策略优化显存与性能
- 保障训练稳定性，分析 loss spike 根因并实现秒级恢复及慢节点自动剔除
Culture
Autonomous TeamsFast-Paced
Requirements
Regions: Us
Get jobs like this in your inbox
Weekly hiring trends and salary data — free.
Join 8 engineers getting weekly insights
Get job market intelligence in your inbox
Free weekly insights on tech hiring trends, salaries, and in-demand stacks.
Already a subscriber? Sign in
About embedding-vc
Industry: saas
Size: startup
Lingopal builds real-time AI language technology and is developing an AI agent platform for enterprises with heavy customer-support needs.
View company profile →Similar Jobs
视频数据标注 & 评测专家 / 负责人
embedding-vc · San Francisco Bay Area
Member of ML Technical Staff
pragmatike · San Francisco
$200k
Member of Technical Staff - Pre-Training Infra
Reflection AI · San Francisco, CA
Member of Technical Staff — Pretraining Infra (Experienced)
Nuance Communications · Seattle, Washington
$300k
Member of Technical Staff - Mid-Training Infra
Reflection AI · San Francisco, CA