AIPaperNotes

Record daily reading of papers and related reproduction results in Chinese.

For more notes, please follow the blog: https://nopsled.blog.csdn.net/

Paper Catalog

Architecture

• Diffusion:
  • DiT
• Flow:
  • Flow Matching [Link]
• MOE:
  • Switch Transformer [Link]
  • DeepSeekMoE
  • Loss-Free Balancing [Link]
• Vision Transformer:
  • ViT [Link]
  • NaViT [Link]

Embedding

• NV-EMBED [Link]
• Qwen3 Embedding [Link]

LLM

• Agent: LLM-based Single/Multi Agent model/system

  • DeepResearch:
    • WebSailor [Link]
    • WebWatcher [Link]
    • RE-TRAC [Link]
  • Memory:
    • Dynamic Cheatsheet [Link]
    • EgoMem [Link]
    • ReasoningBank [Link]
  • Multi Agent Optimization
    • OWL [Link]
    • Multi-Agent Design [Link]
  • RAG
    • BookRAG [Link]
  • Reflection:
    • Reflexion [Link]
    • Metacognitive Reuse [Link]
  • Router:
    • Router-R1 [Link]
  • Visual Agent
    • PixelCraft [Link]
    • SWE-Vision

• Base Model: Large Language Model

  • DeepSeek
    • DeepSeek-V2 [Link]
    • DeepSeek-V3 [Link]
    • DeepSeek-V3.2 [Link]
  • Moonshot AI
    • KIMI LINEAR [Link]
  • Zhipu AI
    • GLM 4.5 [Link]
    • GLM 5 [Link]
  • OpenAI
    • gpt-oss [Link]

• Dataset: Data building and processing for Model training

  • Pretrain:
    • DELT [Link]
  • SFT:
    • OpenSeeker [Link]

• Long Sequence

  • RLM [Link]

• Prompt: Prompt Engineering

  • Context Learning
    • ACE [Link]
  • Skills
    • Extending Claude’s capabilities with skills and MCP servers [Link]
    • Building agents with Skills: Equipping agents for specialized work [Link]

• Omni: LLM-based full modal model

  • Qwen2.5 - Omni [Link]
  • M3 - Agent [Link]

• Quantization: Model Weight/Optimizer/Activation Compressing

  • COAT [Link]

• Speech: Speech LLM

  • ALM: Audio LLM for auido Input
    • Audio Flamingo 3 [Link]

• Survey

  • DeepRearch [Link]
  • Vibe Coding [Link]

• Training: LLM Model Training:

  • Ptrtrain
    • FIM (fill-in-the-middle) [Link]
  • RL
    • RLHF: Reinforcement Learning from Human Feedback
      • BCO [Link]
    • RLRF: Reinforcement Learning with Rich Feedback
      • SDPO [Link]
    • RLVR: Reinforcement Learning with Verifiable Rewards
      • Deepseek - R1 [Link]
      • Dr.GRPO [Link]
      • DAPO [Link]
      • GCG [Link]
      • LUFFY [Link]
      • GSPO [Link]
      • DeepSeek - R1 v2 [Link]
  • SFT:
    • EAFT
  • Speculative Decoding or MTP: Speculative Decoding or Multi-token Prediction
    • Better & Faster Large Language Models via Multi-token Prediction [Link]
    • CAFT [Link]
    • EAGLE3 [Link]

• VLM: Visual LLM

  • LLaVA [Link]
  • Qwen - VL [Link]
  • Qwen2 - VL [Link]
  • Qwen2.5 - VL [Link]
  • Qwen3 - VL [Link]
  • MiniCPM-V 4.5 [Link]
  • DeepSeek - OCR [Link]
  • DeepSeek - OCR2
  • Kimi K2.5 [Link]

Visual Encoder

• Image Segment Pretraining

  • SAM [Link]
  • SAM2 [Link]

• Language-Image Representation Learning:

  • CLIP [Link]
  • SigLIP [Link]
  • SigLIP2 [Link]
  • LIFT [Link]