M.S. student @ Fudan University, School of Data Science, advised by Prof. Baojian Zhou · Knowledge Works Lab
Research Intern @ Alibaba — AI4S / Multimodal LLM
Research Interests: Large Language Models (Agents), Diffusion Language Models, AI for Science, Multimodal LLMs, Graph Neural Networks
| Paper | Venue | Role |
|---|---|---|
| Locality-aware Diffusion Language Modeling — Scatter & Jigsaw blockwise architectures bridging AR and Diffusion | Preprint | Sole 1st Author |
| STM: Spatio-Temporal Distance Model for Dynamic Graph Fraud Detection — SOTA + invention patent | CCF-B | 1st Author |
| SemDLM: Semantic Diffusion Language Modeling — 27.19 Test PPL on LM1B | - | 2nd Author |
Interview prep notes for LLM / Multimodal / RL. Feel free to star and use!
| Resource | Description |
|---|---|
| Agent Harness 深度解析 | Agent 训练基础设施全景 — 四层架构、异步 Rollout、GRPO、Reward 工程、VERL/ROLL/RAGEN 框架对比 |
| LLM 算法岗面试题库 (420+ 题) | Transformer / RLHF / RL / Agentic RL / VLM / Agent / RAG / Infra / 手撕代码，含详解答案 |
| LLM & Multimodal Interview Notes | Transformer, LLM Architecture, RLHF/DPO, Multimodal (CLIP/LLaVA/GPT-4o), Engineering |
| RL for LLM Alignment | Policy Gradient → PPO → GRPO, DQN/DDPG/TD3/SAC, RLHF Pipeline, DPO Derivation, RLVR |
| VLM Knowledge & Interview (2025-2026) | Visual Encoder, VLM Architecture, Alignment, Resolution, MoE |
| Transformer Decoder 全流程 | 训练(并行) vs 推理(自回归+KV Cache)、张量维度推导、RoPE/Attention/MLP 维度链 |
| Training Infra & Distributed Systems | DDP/FSDP/ZeRO/TP/PP/3D Parallel, LoRA/QLoRA, Flash Attention 1-3, Mixed Precision |
| Inference Optimization & System Design | KV Cache, PagedAttention, Quantization, Speculative Decoding, System Design |
| NLP & LLM Course Notes | Tokenization, N-gram, Transformer, GPT, BERT, RLHF (Fudan CS40008) |
Last updated: May 2026