Posts

后端部署第二步：本地到公网——如何使用 Nginx 发布 FastAPI 服务 # 在当今的开发环境中，快速构建和部署后端服务变得至关重要。FastAPI 作为一个高性能、现代化的 Python 异步 Web 框架，广受开发者喜爱。而 Nginx 则是部署 Web 应用最常见也是最稳定的解决方案之一。

Fast-dLLM：通过 KV Cache 和并行 Decoding 加速 dLLM # Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding

后端部署第一步：Nginx 发布服务前的准备工作 # 你是不是刚刚用 Java、Python 或 Go 写出了第一个后端程序？它在本地跑得飞快，通过 localhost:8080 就能访问，功能也都挺顺畅。接下来，自然而然会冒出一个念头：“我要怎样才能让别人也能访问到它？”

自信地深度思考 # DEEP THINK WITH CONFIDENCE

TriangleMix：无损且高效的用于预填充阶段的注意力模式 # TriangleMix: A Lossless and Efficient Attention Pattern for Long Context Prefilling

ASC：CoT 压缩的激活引导 Training free # Activation Steering for Chain-of-Thought Compression

SEAL：大语言模型的可操控推理 Traning Free # SEAL: Steerable Reasoning Calibration of Large Language Models for Free

Hmm 等 Token 影响模型推理能力 # Demystifying Reasoning Dynamics with Mutual Information: Thinking Tokens are Information Peaks in LLM Reasoning

AdaCoT：通过强化学习实现的帕累托最优自适应链式思维触发器 # AdaCoT: Pareto-Optimal Adaptive Chain-of-Thought Triggering via Reinforcement Learning

AdaptThink: 让模型决定是否思考 # AdaptThink: Reasoning Models Can Learn When to Think

SpecReason：使用推测性推理实现加速推理 # SpecReason: Fast and Accurate Inference-Time Compute via Speculative Reasoning

ThinkLess：一种无需训练的推理高效方法，用于减少推理冗余 # ThinkLess: A Training-Free Inference-Efficient Method for Reducing Reasoning Redundancy

Thinkless: LLM Learns When to Think # Thinkless: LLM Learns When to Think

🧠 思维操控：外部 CoT 辅助大模型推理 # Thought Manipulation: External Thought Can Be Efficient for Large Reasoning Models

⏰ 开始作答：弹性推理实现的可扩展的 CoT # Scalable Chain of Thoughts via Elastic Reasoning

DEER：基于 Trial 置信度的推理早停 # DYNAMIC EARLY EXIT IN REASONING MODELS

SoftCoT：Prompt➡️SLM➡️LLM # SoftCoT: Soft Chain-of-Thought for Efficient Reasoning with LLMs

CODI: 通过自蒸馏将 CoT 压缩到连续空间中 # CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillation

HAWKEYE：大小模型协作实现精简 CoT # Hawkeye:Efficient Reasoning with Model Collaboration

LCIRC: 长文档循环压缩方法 （LLM Training Free） # LCIRC: A Recurrent Compression Approach for Efficient Long-form Context and Query Dependent Modeling in LLMs

LLM 在连续 Latent 空间中推理 # Training Large Language Models to Reason in a Continuous Latent Space

Pause Token：隐式 CoT # Think before you speak: Training Language Models With Pause Tokens

面向模型推理思考优化的 Test time scaling # Towards Thinking-Optimal Scaling of Test-Time Compute for LLM Reasoning

CCoT：通过密集表示实现高效推理 # Compressed Chain of Thought: Efficient Reasoning Through Dense Representations

Token 预算意识的 llm 推理 # Token-Budget-Aware LLM Reasoning

INFTYTHINK：打破大型语言模型长上下文推理长度限制 # InftyThink: Breaking the Length Limits of Long-Context Reasoning in Large Language Models

TokenSkip:可控的 CoT 压缩 in LLMs # TokenSkip: Controllable Chain-of-Thought Compression in LLMs

LightThinker: 每个想法压缩成两个 token # 如果 lt 确实靠谱