ai算法工程师（后训练）.

• Design, provision, and operate large-scale distributed training environments across multi-GPU, multi-node clusters using Kubernetes and managed ML platforms (AWS SageMaker HyperPod/EKS, Tencent TiOne, etc.).

• Configure and tune model/data parallelism strategies (data, tensor, pipeline, sequence, MoE) to maximize throughput.

• Improve training efficiency, manage storage pipelines (S3, EFS, FSx for Lustre), and engineer robust checkpointing/fault tolerance.

• Build observability tools, CI/CD pipelines, and infrastructure-as-code for training.

• Design and run post-training workflows (SFT, RFT/RLVR/RLAIF, continual pretraining) and integrate expert models into production drug-discovery workflows.

skills and experience required.

• Education: Bachelor in Computer Science and above.

• Experience Level: 5+ years of hands-on experience spanning ML infrastructure / LLMOps and LLM training.

• Technical Skills:

  • Deep, hands-on expertise with Kubernetes for large-model training across multiple platforms (AWS, GCP, Azure, or Tencent TiOne).

  • Strong understanding of distributed-training parallelism strategies.

  • Hands-on experience with frameworks like PyTorch (DDP/FSDP), DeepSpeed/ZeRO, and the Hugging Face ecosystem.

  • Fluency with cloud storage and high-throughput data systems (S3, EFS, FSx for Lustre).

  • Practical understanding of post-training and RL methods (SFT, GRPO, PPO, DPO, reward modeling).

  • Strong Python, container/orchestration (Docker/Kubernetes), and infrastructure-as-code skills.

• Desirable Extras (Nice-to-Have):

  • Experience with Megatron-LM, NVIDIA NeMo, TRL, verl, or vLLM.

  • Knowledge of cluster networking (NCCL tuning, EFA/RDMA/InfiniBand) and schedulers (Slurm, Ray, Volcano).

  • Familiarity with AI-for-Science applications (biomedical, chemical) and experiment tracking tools (W&B, MLflow).