RSI：Recursive Self-Improvement 分类图谱

GRPO训练排查：策略坍塌 vs 训练崩溃

NoPE

LLM 算法岗面试手册 2026

从 GPT-2 到 Kimi K3：大模型架构如何演进

MOPD：Multi-Teacher On-Policy Distillation for Capability Integration in LLM Post-Training

verl V1 主线重构 · 给后训练工作者的三问

Review：自进化Agent的RL Memory与Skill方法批判

DeepSeek V4 Infra Deep Dive 4.4: PyTorch 训练显存管理

RL 面试题 2026

The Ultra-Scale Playbook 学习笔记

滑动窗口注意力（Sliding Window Attention）

Xiaomi MiMo-V2-Flash

SmolLM3

Qwen3 系列（Qwen3 / Qwen3-Next / Coder-Next）

QK-Norm

Post-Norm vs Pre-Norm

Partial RoPE

OLMo 2 & Olmo 3

NoPE（No Positional Embedding）

Nemotron 3（Nano / Super）

MTP（Multi-Token Prediction）

MoE（Mixture of Experts）

MLA（Multi-Head Latent Attention）

Mistral 系列（Small 3.1 / Mistral 3）

MiniMax M2

Llama 4

Latent Experts

Kimi 系列（K2 / K2 Thinking / Kimi Linear）

KDA（Kimi Delta Attention）

Grok 2.5

GPT-OSS（OpenAI 的开源模型）

GLM 系列（GLM-4.5 / GLM-5）

Gemma 系列（Gemma 3 / 3n / 4）

Gated DeltaNet

Feb 2026 LLM 新发布（速览）

DeepSeek V3 系列

Attention Sinks

Arcee AI Trinity Large

2025-2026 LLM 架构对比

Qwen2.5-3.7 系列最全总结

veRL中colocate实现源码解析（二）

veRL基于Ray的执行流程源码解析（一）

nano-vllm 全代码精读笔记:用 1200 行 Python 理解 vLLM 推理引擎

Claude Code Hooks 笔记

SAPO

CISPO

cc常用命令

vllm高效推理全链路

一些参数

ai阅读megatron

hf2megatron

training_model

llavaov_1_5

cr阅读记录

qwen3vl+deepseek

patch详解

快速入门

metrics总结

自创容器

example

ReasoningBank：Scaling Agent Self-Evolving with Reasoning Memor

GVPO：Group Variance Policy Optimization for Large Language Model Post-Training

agentic RL

Skywork-R1V3 Technical Report

ToolRM：Outcome Reward Models for Tool-Calling Large Language Models

Tool-R1：Sample-Efficient Reinforcement Learning for Agentic Tool Use

Single-stream Policy Optimization

Process-Supervised Reinforcement Learning for Interactive Multimodal Tool-Use Agents

H2R：Hierarchical Hindsight Reflection for Multi-Task LLM Agents

Disentangling Memory and Reasoning Ability in Large Language Models

DeepDive：Advancing Deep Search Agents with Knowledge Graphs and Multi-Turn RL

AGENTRACER：WHO IS INDUCING FAILURE IN THE LLM AGENTIC SYSTEMS

Empowering LLMs with Parameterized Skills for Adversarial Long-Horizon Planning

The Illusion of Diminishing Returns：Measuring Long Horizon Execution in LLMs

Harnessing Uncertainty：Entropy-Modulated Policy Gradients for Long-Horizon LLM Agents

Bridging the Capability Gap：Harmonizing Multi-Agent Systems via Joint Alignment Tuning

ZEP：A TEMPORAL KNOWLEDGE GRAPH ARCHITECTURE FOR AGENT MEMORY

RAGEN：Understanding Self-Evolution in LLM Agents via Multi-Turn Reinforcement Learning

Parallel-R1：Towards Parallel Thinking via Reinforcement Learning

K2-Think：A Parameter-Efficient Reasoning System

Exploratory Retrieval-Augmented Planning For Continual Embodied Instruction Following

EnvX：Agentize Everything with Agentic AI

AgentGym-RL：Training LLM Agents for Long-Horizon Decision Making through Multi-Turn Reinforcement Learning

WebExplorer：Explore and Evolve for Training Long-Horizon Web Agents

The Majority is not always right：RL training for solution aggregation

TableMind：An Autonomous Programmatic Agent for Tool-Augmented Table Reasoning

SFR-DeepResearch：Towards Effective Reinforcement Learning for Autonomously Reasoning Single Agents

SELF-ALIGNED REWARD：TOWARDS EFFECTIVE AND EFFICIENT REASONERS

Reverse-Engineered Reasoning for Open-Ended Generation

ReSearch：Learning to Reason with Search for LLMs via Reinforcement Learning

Plan Verification for LLM-Based Embodied Task Completion Agents

Less is More Tokens：Efficient Math Reasoning via Difficulty-Aware Chain-of-Thought Distillation

AGENT WORKFLOW MEMORY

VERLTOOL：TOWARDS HOLISTIC AGENTIC REINFORCEMENT LEARNING WITH TOOL USE

Planning with Reasoning using Vision Language World Model

Learning When to Plan：Efficiently Allocating Test-Time Compute for LLM Agents

Learning to Refine：Self-Refinement of Parallel Reasoning in LLMs

Design and Optimization of Reinforcement Learning-Based Agents in Text-Based Games

BEYOND CORRECTNESS：HARMONIZING PROCESS AND OUTCOME REWARDS THROUGH RL TRAINING

未命名

SimpleTIR：End-to-End Reinforcement Learning for Multi-Turn Tool-Integrated Reasoning

DCPO：DYNAMIC CLIPPING POLICY OPTIMIZATION

Think in Games：Learning to Reason in Games via Reinforcement Learning with Large Language Models

rStar2-Agent：Agentic Reasoning Technical Report

Understanding Tool-Integrated Reasoning

MUA-RL：MULTI-TURN USER-INTERACTING AGENT REINFORCEMENT LEARNING FOR AGENTIC TOOL USE

GHPO：ADAPTIVE GUIDANCE FOR STABLE AND EFFICIENT LLM REINFORCEMENT LEARNING

AI-SEARCHPLANNER：MODULAR AGENTIC SEARCH VIA PARETO-OPTIMAL MULTI-OBJECTIVE REINFORCEMENT LEARNING

ThinkDial：An Open Recipe for Controlling Reasoning Effort in Large Language Models

Memory-R1：Enhancing Large Language Model Agents to Manage and Utilize Memories via Reinforcement Learning

Memento：Fine-tuning LLM Agents without Fine-tuning LLMs

MCPVerse：An Expansive, Real-World Benchmark for Agentic Tool Use

HIPLAN：Hierarchical Planning for LLM Agents with Adaptive Global-Local Guidance

CARFT：Boosting LLM Reasoning via Contrastive Learning with Annotated Chain-of-Thought-based Reinforced Fine-Tuning

memory论文整理

From Trial-and-Error to Improvement：A Systematic Analysis of LLM Exploration Mechanisms in RLVR

Coarse-to-Fine Grounded Memory for LLM Agent Planning

Chain-of-Agents：End-to-End Agent Foundation Models via Multi-Agent Distillation and Agentic RL

未命名

Your Reward Function for RL is Your Best PRM for Search：Unifying RL and Search-Based TTS

Think in Blocks：Adaptive Reasoning from Direct Response to Deep Reasoning

Self-Guided Function Calling in Large Language Models via Stepwise Experience Recall

Reinforcement Learning with Rubric Anchors

MemAgent：Reshaping Long-Context LLM with Multi-Conv RL-based Memory Agent

CausalPlan：Empowering Efficient LLM Multi-Agent Collaboration Through Causality-Driven Planning

Beyond the 80 20 Rule：High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning

BEYOND PASS@1：SELF-PLAY WITH VARIATIONAL PROBLEM SYNTHESIS SUSTAINS RLVR

Small Language Models are the Future of Agentic AI

MemTool：Optimizing Short-Term Memory Management for Dynamic Tool Calling in LLM Agent Multi-Turn Conversations

ETTRL：BALANCING EXPLORATION AND EXPLOITATION IN LLM TEST-TIME REINFORCEMENT LEARNING VIA ENTROPY MECHANISM

EDGE-GRPO：Entropy-Driven GRPO with Guided Error Correction for Advantage Diversity

CURE：Critical-Token-Guided Re-concatenation for Entropy-collapse Prevention

context_engineering

SSRL：SELF-SEARCH REINFORCEMENT LEARNING

Pruning Long Chain-of-Thought of Large Reasoning Models via Small-Scale Preference Optimization

Promoting Efficient Reasoning with Verifiable Stepwise Reward

TURA：Tool-Augmented Unified Retrieval Agent for AI Search

Sample More to Think Less：Group Filtered Policy Optimization for Concise Reasoning

ParallelSearch：Train your LLMs to Decompose Query and Search Sub-queries in Parallel with Reinforcement Learning

Beyond Ten Turns：Unlocking Long-Horizon Agentic Search with Large-Scale Asynchronous RL

AWORLD：DYNAMIC MULTI-AGENT SYSTEM WITH STABLE MANEUVERING FOR ROBUST GAIA PROBLEM SOLVING

Agent Lightning：Train ANY AI Agents with Reinforcement Learning

TRAIN LONG, THINK SHORT：CURRICULUM LEARNING FOR EFFICIENT REASONING

分层规划相关论文

Memp：Exploring Agent Procedural Memory

Enhancing LLM-Based Agents via Global Planning and Hierarchical Execution

Empowering Large Language Model Agents through Action Learning

R-Zero：Self-Evolving Reasoning LLM from Zero Data

Group Causal Policy Optimization for Post-Training Large Language Models

qwen

GTPO：Trajectory-Based Policy Optimization in Large Language Models

GTPO and GRPO-S：Token and Sequence-Level Reward Shaping with Policy Entropy

agent RL相关论文

Training Long-Context, Multi-Turn Software Engineering Agents with Reinforcement Learning

PilotRL：Training Language Model Agents via Global Planning-Guided Progressive Reinforcement Learning

Group-in-Group Policy Optimization for LLM Agent Training

Efficient Agents：Building Effective Agents While Reducing Cost

CAPO：Towards Enhancing LLM Reasoning through Verifiable Generative Credit Assignment

Beyond Policy Optimization：A Data Curation Flywheel for Sparse-Reward Long-Horizon Planning

信用分配

SIMURA：Towards General Goal-Oriented Agent via Simulative Reasoning Architecture with LLM-Based World Model

GEPA：REFLECTIVE PROMPT EVOLUTION CAN OUTPERFORM REINFORCEMENT LEARNING

Geometric-Mean Policy Optimization

LAPO：INTERNALIZING REASONING EFFICIENCY VIA LENGTH-ADAPTIVE POLICY OPTIMIZATION

Thinking vs. Doing：Agents that Reason by Scaling Test-Time Interaction

RLVMR：Reinforcement Learning with Verifiable Meta-Reasoning Rewards for Robust Long-Horizon Agents

ProRL：Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models

Good Learners Think Their Thinking：Generative PRM Makes Large Reasoning Model More Efficient Math Learne

UloRL：An Ultra-Long Output Reinforcement Learning Approach for Advancing Large Language Models’ Reasoning Abilities

K2

AGENTIC REINFORCED POLICY OPTIMIZATION

VAPO：Efficient and Reliable Reinforcement Learning for Advanced Reasoning Tasks

Gemma

ALiBi

Group Sequence Policy Optimization

Towards Effective Code-Integrated Reasoning

device_mesh

Dataset

Routine：A Structural Planning Framework for LLM Agent System in Enterprise

agent_loop

tensor_parallel

pipeline parallelism

data_parallel

zero

Search and Refine During Think：Autonomous Retrieval - Augmented Reasoning of LLMs

Peri-LN：Revisiting Normalization Layer in the Transformer Architecture

flops分析

ZEROSEARCH：Incentivize the Search Capability of LLMs without Searching

ring-all-reduce

WebThinker：Empowering Large Reasoning Models with Deep Research Capability

Search-R1：Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning

PROCESS REINFORCEMENT THROUGH IMPLICIT REWARDS

MQA

MHA

BRiTE：Bootstrapping Reinforced Thinking Process to Enhance Language Model Reasoning

batch_size 解释

RLOO

ReMAX（REINFORCE argmax）

REINFORECE++

Reinforcing General Reasoning without Verifiers

ppo

paged attention

online attention

grpo

GQA

flash attention

First Return, Entropy-Eliciting Explore

dapo

SGD-Momentum

SGD

RMSProp

AdamW

AdaGrad

Muon

MLA

梯度累计

RLPR：EXTRAPOLATING RLVR TO GENERAL DOMAINS WITHOUT VERIFIERS

GENERALIST REWARD MODELS：FOUND INSIDE LARGE LANGUAGE MODELS

CPU offloading

Activation checkpointing

DataProto

ulysses_sequence_parallel

remove_padding

LAN-AND-ACT：Improving Planning of Agents for Long-Horizon Tasks

长度扩展（外推）

显存占用计算

思维链压缩

transformer参数量分析

fsdp

entropy(reasoning)

WebEvolver：Enhancing Web Agent Self-Improvement with Coevolving World Model

WEB AGENTS WITH WORLD MODELS ：LEARNING AND LEVERAGING ENVIRONMENT DYNAMICS IN WEB NAVIGATION

world_model

ray前置知识

init_workers详解

RLVR-World

agent概览

verl总体概览

reward_mananger

MCTS和PRM

LEGB

debugger

generate

repeat

reduce

rearrange

pack and unpack

einsum

parse_shape

asnumpy

gather和scatter

agent planning综述

llama系列

frequency_penalty&presence_penalty

rwkv

rope

Data Engineering for Scaling Language Models to 128K Context

Transformer Feed-Forward Layers Are Key-Value Memories

MoE

Lora微调

KV cache

课程表（拓扑排序）

跳跃游戏

乘积最大的数组

缺失的第一个正数

和为K的子数组

和为K的子数组

并查集

T5

温度超参数

验证外星语词典

FP-Growth

Layer Norm

L1 L2正则化

Batch Norm

hinge loss

北京百分点面经

特征选择

KMP

shortcode(置顶)

最大熵模型

focal loss

early-stopping

标签平滑

warmup

数据不平衡

对抗训练

调参技巧

xgboost

GAN

LDA

GRU

bincount

把数字翻译成字符串

seq2seq

DBSCAN

关联规则概念

字符串转换整数 (atoi)

高斯混合聚类

tokenization

CoVe

分发饼干

EM算法

regex

aidc测试

Adam算法

决策树

精确率和召回率

分类算法概述

BM25

GPT

最长回文子串

Dropout正则化

有效的数独

Apriori算法

使括号有效的最少添加

Ensemble Learning

旋转图像

KNN

主题模型

最大子序和

Transformer

使用最小花费爬楼梯

条件随机场

Adaboost

分割等和子集

kd树

最小公众前缀

树回归

随机森林

搜索旋转排序数组

learn_four

线性判别分析

bayes

rot90

数据挖掘比赛

前缀树

Logistic回归

melt

预训练模型

flip

ELMo

线性回归

Stacking

计数质数

感知机算法

最大数

LSTM

检查平衡性

检查平衡性

语言模型

HMM

实训学习内容

TF-IDF

平衡括号字符串的最少插入次数

WTF

共现矩阵

skill

长度最小的子数组

SVM

字符串的排列

apply_along_axis

采购方案

只出现一次的数字

谱聚类

概率图模型概述

B树

和为s的连续正数序列

合并区间

三数之和

Word Embedding

SVD

层次聚类

ROC曲线

删除排序数组中的重复项

asyncio

移除元素

TextRank

pyplot

机器学习的一些概念

交叉熵损失函数

验证二叉搜索树

RNN

vim

Smooth L1 Loss

无重叠区间

对角线遍历

PCA

BERT

git

编辑距离

kmeans

翻转二叉树

对称二叉树

合并两个有序列表

可获得的最大点数

滑动窗口中位数

Attention

PPMI

至少有k个重复字符的最长字串

最长递增子序列

外观数列

as_strided

NER

阶乘函数后K个零(首个困难题)

打家劫舍

thread

丑数系列

GBDT

两两交换链表中的节点

种花问题（新年快乐!2021第一题）

零钱兑换

去除重复字母