Alex Xi's Notes

Tag: architecture

18 items with this tag.

  • Jul 06, 2026

    tweet-sparse-attention-anns

    • attention
    • sparse-attention
    • anns
    • retrieval
    • architecture
  • Jul 06, 2026

    tweet-subq-sub-quadratic-llm

    • architecture
    • attention
    • long-context
    • efficiency
  • Jul 06, 2026

    tweet-anti-moe-take

    • moe
    • architecture
    • contrarian
  • Jul 06, 2026

    tweet-attention-free-model-14b

    • architecture
    • attention-free
    • transformer
    • alternative
    • efficiency
  • Jul 06, 2026

    tweet-causal-jepa-icml

    • jepa
    • self-supervised-learning
    • icml
    • architecture
  • Jul 06, 2026

    tweet-deepseek-v4-analysis

    • deepseek
    • architecture
    • moe
    • scaling
  • Jul 06, 2026

    tweet-deepseek-v4-hbm

    • deepseek
    • architecture
    • hbm
    • hardware
  • Jul 06, 2026

    tweet-gemma4-architecture

    • architecture
    • gemma
    • google
    • transformer
    • model-design
  • Jul 06, 2026

    tweet-group-representational-pos-encoding

    • positional-encoding
    • architecture
    • jane-street
  • Jul 06, 2026

    tweet-moe-training-instability

    • moe
    • training-stability
    • architecture
  • Jul 06, 2026

    tweet-parallel-agent-context-search

    • agents
    • parallelism
    • context
    • search
    • architecture
  • Jul 06, 2026

    mixture-experts-architectures

    • mixture-of-experts
    • moe
    • architecture
    • scaling
  • Jul 06, 2026

    openai-reverse-engineered

    • openai
    • o1
    • architecture
    • reverse-engineering
  • Jul 06, 2026

    recursive-language-models-2026

    • recursive-models
    • training
    • architecture
    • self-improvement
  • Jul 06, 2026

    differential-transformer

    • differential-attention
    • transformer
    • architecture
    • attention
  • Jul 06, 2026

    how-attention-residuals-work

    • transformers
    • attention
    • architecture
  • Jul 06, 2026

    attention-wrong-norm

    • attention
    • transformers
    • architecture
    • normalization
    • sequence-length
  • Jul 06, 2026

    rag-vs-compiled-knowledge

    • rag
    • knowledge-management
    • llm
    • architecture