Natural Language Processing

193페이지

[논문리뷰]: NVIDIA Nemotron 3: Efficient and Open Intelligence

논문 링크 NVIDIA Nemotron 3: 하이브리드 Mamba–Transformer MoE로 “정확도/처리량” 프런티어를 당기다Nemotron 3 는 MoE 하이브리드 Mamba–Transformer, LatentMoE , MTP , NVFP4 학 …

2025년 12월 16일

4분

Natural Language Processing Machine Learning Mixture of Experts

[논문리뷰] Pretraining Large Language Models with NVFP4

논문 링크 NVFP4로 4-bit 프리트레이닝을 실전으로: 12B를 10T 토큰까지, FP8과 사실상 동급TL;DR12B 하이브리드 Mamba-Transformer를 10T tokens 에서 NVFP4(4-bit) 로 프리트레이닝하면 안정 구간 손실차 …

2025년 10월 9일

39분

Natural Language Processing Machine Learning

[논문리뷰] Inference-Time Hyper-Scaling with KV Cache Compression

논문 링크 Dynamic Memory Sparsification(DMS): KV 캐시 8× 압축으로 LLM 하이퍼-스케일링을 현실로한 줄 요약 (TL;DR)1 K 스텝만의 경량 재적합과 지연 퇴출 전략을 결합한 DMS는 KV 캐시를 최대 8× 압축하면 …

2025년 7월 29일

29분

Machine Learning Natural Language Processing KV Cache

Hogwild! Inference: Parallel LLM Generation via Concurrent Attention

논문 링크 Q : 이 논문을 아주 자세하게 읽고 논문의 강점과 독창적인 지점을 설명해주고 핵심 알고리즘을 예시 입력을 들어서 전체적인 과정을 설명해줘 추가적으로 논문의 한계점에 대해서도 알려줘Hogwild! Inference 논문 심층 분석제공해주신 …

2025년 6월 19일

38분

Machine Learning Natural Language Processing

Towards Economical Inference: Enabling DeepSeek’s Multi-Head Latent Attention in Any Transformer-based LLMs

논문 링크 Q : 이 논문을 아주 자세하게 읽고 논문의 강점과 독창적인 지점을 설명해주고 핵심 알고리즘을 예시 입력을 들어서 전체적인 과정을 설명해줘 추가적으로 논문의 한계점에 대해서도 알려줘MHA2MLA 논문 분석: 기존 LLM을 경제적으로 만드는 …

[논문리뷰]: NVIDIA Nemotron 3: Efficient and Open Intelligence

[논문리뷰] Pretraining Large Language Models with NVFP4

[논문리뷰] Inference-Time Hyper-Scaling with KV Cache Compression

Hogwild! Inference: Parallel LLM Generation via Concurrent Attention

Towards Economical Inference: Enabling DeepSeek’s Multi-Head Latent Attention in Any Transformer-based LLMs

X-EcoMLA: Upcycling Pre-Trained Attention into MLA for Efficient and Extreme KV Compression

XGRAMMAR: FLEXIBLE AND EFFICIENT STRUCTURED GENERATION ENGINE FOR LARGE LANGUAGE MODELS

RODIMUS*: BREAKING THE ACCURACY-EFFICIENCY TRADE-OFF WITH EFFICIENT ATTENTIONS

Gemini Embedding: Generalizable Embeddings from Gemini

Gemma 3 Technical Report

SwitchHead: Accelerating Transformers with Mixture-of-Experts Attention

SparseTransX: Efficient Training of Translation-Based Knowledge Graph Embeddings Using Sparse Matrix Operations

SELF-DATA DISTILLATION FOR RECOVERING QUALITY IN PRUNED LARGE LANGUAGE MODELS

SampleAttention: Near-Lossless Acceleration of Long Context LLM Inference with Adaptive Structured Sparse Attention

QServe: W4A8KV4 Quantization and System Co-design for Efficient LLM Serving

EFFICIENT LLM INFERENCE USING DYNAMIC INPUT PRUNING AND CACHE-AWARE MASKING

Comet: Fine-grained Computation-communication Overlapping for Mixture-of-Experts

LSERVE: EFFICIENT LONG-SEQUENCE LLM SERVING WITH UNIFIED SPARSE ATTENTION

Speculate, then Collaborate: Fusing Knowledge of Language Models during Decoding

You OnlyPruneOnce: DESIGNING CALIBRATION-FREE MODEL COMPRESSION WITH POLICY LEARNING

TypedThinker: Typed Thinking Improves Large Language Model Reasoning

LASP-2: Rethinking Sequence Parallelism for Linear Attention and Its Hybrid

Robust and Secure Code Watermarking for Large Language Models via ML/Crypto Codesign

SmolLM2: When Smol Goes Big Data-Centric Training of a Small Language Mode

DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

DeepSeek-VL2: Mixture-of-Experts Vision-Language Models for Advanced Multimodal Understanding

Humanity's Last Exam

Qwen2.5-1M Technical Report

Qwen2-VL: Enhancing Vision-Language Model’s Perception of the World at AnyResolution

DeepSeek-Prover-V1.5: Harnessing Proof Assistant Feedback for Reinforcement Learning and Monte-Carlo Tree Search

Let the Expert Stick to His Last: Expert-Specialized Fine-Tuning for Sparse Architectural Large Language Models

Qwen2 Technical Report

How to Train Data-Efficient LLMs

DeepSeek-Coder: When the Large Language Model Meets Programming - The Rise of Code Intelligence

DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

DeepSeek LLM: Scaling Open-Source Language Models with Longtermism

Qwen-Audio: Advancing Universal Audio Understanding via Unified Large-Scale Audio-Language Models

Qwen Technical Report

Qwen-VL: AVersatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond

Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation

Compressed Context Memory For Online Language Model Interaction

Qwen2.5 Technical Report

AIOS: LLM Agent Operating System

DynamicKV: Task-Aware Adaptive KV Cache Compression for Long Context LLMs

SeerAttention: Learning Intrinsic Sparse Attention in Your LLMs

TAIPAN: EFFICIENT AND EXPRESSIVE STATE SPACE LANGUAGE MODELS WITH SELECTIVE ATTENTION

Block Transformer: Global-to-Local Language Modeling for Fast Inference

FLAME: Factuality-Aware Alignment for Large Language Models

MobiLlama: Towards Accurate and Lightweight Fully Transparent GPT

Rethinking Optimization and Architecture for Tiny Language Models

Cascade Speculative Drafting for Even Faster LLM Inference

Gated Linear Attention Transformers with Hardware-Efficient Training

LLM in a flash : Efficient Large Language Model Inference with Limited Memory

Agile-Quant: Activation-Guided Quantization for Faster Inference of LLMs on the Edge

Improving alignment of dialogue agents via targeted human judgements

Language Models are General-Purpose Interfaces

OPT: Open Pre-trained Transformer Language Models

CBQ: Cross-Block Quantization for Large Language Models

SCCA: Shifted Cross Chunk Attention for long contextual semantic expansion

Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context

Gemma: Open Models Based on Gemini Research and Technology

LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding

The Truth is in There: Improving Reasoning in Language Models with Layer-Selective Rank Reduction

Gemma 2: Improving Open Language Models at a Practical Size

The Llama 3 Herd of Models

Communication Compression for Tensor Parallel LLM Inference

Large Concept Models: Language Modeling in a Sentence Representation Space

Sharing and Throughput-oriented Token Batching

Star Attention: Efficient LLM Inference over Long Sequences

Byte Latent Transformer: Patches Scale Better Than Tokens

GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding

Memory Layers at Scale

Orca: Progressive Learning from Complex Explanation Traces of GPT-4

Fast and Effective Weight Update for Pruned Large Language Models

FFSplit: Split Feed-Forward Network For Optimizing Accuracy-Efficiency Trade-off in Language Model Inference

Lightning Attention-2: A Free Lunch for Handling Unlimited Sequence Lengths in Large Language Models

MoE-Mamba: Efficient Selective State Space Models with Mixture of Experts

INFERFLOW: AN EFFICIENT AND HIGHLY CONFIG URABLE INFERENCE ENGINE FOR LARGE LANGUAGE MODELS

MEDUSA: Simple LLMInference Acceleration Framework with Multiple Decoding Heads

Break the Sequential Dependency of LLM Inference Using LOOKAHEAD DECODING