전체 글 썸네일형 리스트형 Ubuntu 24.04.01 LTS 기준 크롬 원격설정 1-1. 크롬 설치를 위한 환경설치sudo apt-get update &&sudo apt-get install -f &&sudo apt-get install xvfb xserver-xorg-video-dummy xbase-clients libutempter0 1-2. 크롬설치sudo dpkg -i chrome-remote-desktop_current_amd64.deb 2. 편한 설정을 위한 gedit 설치sudo apt install -y gedit 3. 원본 파일 들어가기sudo gedit /opt/google/chrome-remote-desktop/chrome-remote-desktop 4-1. FIRST_X_DISPLAY_NUMBER 확인 및 수정printenv DISPLAY이게 :0이라면FIRST.. 더보기 QeRL: Beyond Efficiency -- Quantization-enhanced Reinforcement Learning for LLMs https://arxiv.org/abs/2510.11696 QeRL: Beyond Efficiency -- Quantization-enhanced Reinforcement Learning for LLMsWe propose QeRL, a Quantization-enhanced Reinforcement Learning framework for large language models (LLMs). While RL is essential for LLMs' reasoning capabilities, it is resource-intensive, requiring substantial GPU memory and long rollout durations. QeRLarxiv.org 초록(Abstract)본 논문에서는 .. 더보기 Don't Blind Your VLA: Aligning Visual Representations for OOD Generalization https://arxiv.org/abs/2510.25616 Don't Blind Your VLA: Aligning Visual Representations for OOD GeneralizationThe growing success of Vision-Language-Action (VLA) models stems from the promise that pretrained Vision-Language Models (VLMs) can endow agents with transferable world knowledge and vision-language (VL) grounding, laying a foundation for action models witarxiv.org 초록시각-언어-행동(VLA, Vision-.. 더보기 DoPE: Denoising Rotary Position Embedding https://arxiv.org/abs/2511.09146 DoPE: Denoising Rotary Position EmbeddingRotary Position Embedding (RoPE) in Transformer models has inherent limits that weaken length extrapolation. We reinterpret the attention map with positional encoding as a noisy feature map, and propose Denoising Positional Encoding (DoPE), a training-freearxiv.org 초록Transformer 모델의 Rotary Position Embedding(RoPE)은 구조적으로 길.. 더보기 Quantization, ‘가벼운’ AI 모델 구현을 위한 핵심 기술 https://medium.com/@enerzai/quantization-%EA%B0%80%EB%B2%BC%EC%9A%B4-ai-%EB%AA%A8%EB%8D%B8-%EA%B5%AC%ED%98%84%EC%9D%84-%EC%9C%84%ED%95%9C-%ED%95%B5%EC%8B%AC-%EA%B8%B0%EC%88%A0-b34e21607088 더보기 MXFP8, MXFP4 및 NVFP4에 대한 자세한 설명 https://zhuanlan.zhihu.com/p/1969465397670551963 1. 왜 mxfp8, mxfp4, nvfp4 같은 저정밀도 형식이 필요한가?1) 대규모 모델의 폭발적 성장으로 인한 계산 및 메모리 병목 심화LLM의 파라미터 수는 이미 조 단위에 도달했고, 학습 시 필요한 FLOPs는 10²⁵을 넘는다.기존 FP32나 BF16 형식은 대역폭과 메모리 사용량이 매우 높아, 처리량과 에너지 효율에 제한을 준다.단순히 비트 너비만 줄인 INT8 또는 FP8을 적용하면 동적 범위가 충분하지 않아 학습 발산이나 성능 저하가 발생한다.2) 기존 저정밀도 형식의 구조적 한계INT8: 사전에 scale을 설정해야 하며, LLM의 activation·gradient가 따르는 멱함수(power-law.. 더보기 INT v.s. FP: A Comprehensive Study of Fine-Grained Low-bit Quantization Formats https://arxiv.org/abs/2510.25602 INT v.s. FP: A Comprehensive Study of Fine-Grained Low-bit Quantization FormatsModern AI hardware, such as Nvidia's Blackwell architecture, is increasingly embracing low-precision floating-point (FP) formats to handle the pervasive activation outliers in Large Language Models (LLMs). Despite this industry trend, a unified comparisonarxiv.org 초록최신 AI 하드웨어, 특히 Nvid.. 더보기 Less is More: Recursive Reasoning with Tiny Networks https://arxiv.org/abs/2510.04871?ref=refetch.io Less is More: Recursive Reasoning with Tiny NetworksHierarchical Reasoning Model (HRM) is a novel approach using two small neural networks recursing at different frequencies. This biologically inspired method beats Large Language models (LLMs) on hard puzzle tasks such as Sudoku, Maze, and ARC-AGI while traarxiv.org 초록(Abstract)Hierarchical Reasoni.. 더보기 이전 1 2 3 4 ··· 72 다음