全部文章 35

大模型

7 篇

Omni-Modal

Diffusion Language Models

Chain of Thought

Positional Encoding

DeepSeek AI-I

DeepSeek AI-II

Token Mixer-I

AI Infra

9 篇

Training Infrastructure

Infra Optimizers

FLASH-ATTN

Infra Quant-I

Infra Quant-II

Infra Quant III

Speculative Decoding I

Speculative Decoding II

Speculative Decoding III

策略优化

5 篇

OnPolicy Distill

Policy Optimization-I

Policy Optimization-II

Policy Optimization-III

Policy Optimization-IV

生成模型

6 篇

Generative Models I

Generative Models II

Generative Models III

Generative Models IV

Generative Models V

Generative Models VI

源码解读

4 篇

Gated DeltaNet

Interleaved Rotary Position Embedding

ms-swift ChatML 渲染

FlashAttention Varlen

GitHub Issue

4 篇

ms-swift 模型构造与运行时

Gemma 4 NVFP4 MoE 空输出

ZeRO-3 梯度竞态与 GDN Pipeline 死锁

vLLM #51593：Persistent Top-K CTA 死锁