大模型训练（番茄网文方向）-CQC
Core
Build and optimize large language models for content safety governance and risk assessment in the online novel (web novel) industry.
Role type
Senior IC large language model training engineer (safety alignment)
Builds
Production-ready LLMs for UGC/PGC/AIGC content moderation and compliance generation
Domain
Internet / AI Safety / Content Governance
Deliverable
production ML models
Required skills
Python (Pandas, NumPy), data engineering, prompt engineering, SFT/DPO training, Transformer/MoE architecture knowledge, distributed training (DeepSpeed, FSDP), gradient debugging
Preferred skills
Master's/PhD in CS/AI/Math, experience with Llama/Qwen/DeepSeek models, cluster training operations
Technologies
Python, Pandas, NumPy, Llama, Qwen, DeepSeek, LoRA, DPO, DeepSpeed, FSDP, Linux
Responsibilities
Convert safety standards into computable logic (regex, classification rules) and training data constraints; design and iterate prompts for content governance; execute full model training lifecycle (selection, dataset prep, fine-tuning, evaluation); resolve training errors (gradient vanishing, overfitting, OOM); collaborate with business teams to define boundary cases for safety logic.
Seniority
Senior, hands-on IC
