cool https://github.com/huggingface/candle/pull/3405 fix: conv2d_tiled produces wrong results when C == H == W by developer0hye · Pull Request #3405 · huggingface/candleBug conv2d_tiled produces completely wrong output when C_in == H == W (e.g., input shape [1, 64, 64, 64]). Fixes #3404 Root Cause conv2d_tiled converts input from NCHW to NHWC internally. It uses a...github.com 이거 다 클로드로 한거다. 문제 ..
https://docs.cloud.google.com/vertex-ai/generative-ai/docs/prompt-gallery Generative AI prompt samples | Generative AI on Vertex AI | Google Cloud DocumentationSend feedback Generative AI prompt samples Stay organized with collections Save and categorize content based on your preferences. Choose a sample to view an example of a prompt and a response from one of Google's generative AI models...
공식 hf repo에는 image 한장 처리하는 예제 밖에 없어서 당황했다가 깃헙 리포를 보니 다양한 예제에 대한 코드들이 있었다. https://github.com/QwenLM/Qwen3-VL/tree/main/cookbooks Qwen3-VL/cookbooks at main · QwenLM/Qwen3-VLQwen3-VL is the multimodal large language model series developed by Qwen team, Alibaba Cloud. - QwenLM/Qwen3-VLgithub.com
전반적인 데이터셋에서 높은 성능을 보이는 첫 Realtime End2End Transformer 기반 Object Detection 모델이라 할 수 있는 RF-DETR 논문을 읽어봤다. 실험을 굉장히 많이 했다는 게 느껴지는 논문이다. 회사에서 많이 해왔던 일이라면 공개되는 최신 vision foundation backbone networks들을 우리의 기존 모델의 backbone을 교체하여 성능 변화를 보는 일인데, 이 논문에서 object detection 모델을 대상으로 비교적 최신 vision backbone networks라 할 수 있는 SigLIPv2, SAM2 네트워크에 대해 실험을 해놨다. 실험 결과는 논문에서 Table 6에서 확인할 수 있다. 결론은 DINOv2가 좋다이다. SigLI..
# Role 당신은 해당 분야의 전문 연구원입니다. 주어진 논문을 분석하여 핵심 내용을 명확하고 전문적으로 요약해주세요. # Constraints (Critical) 1. **Technical Terminology in English**: 논문에 등장하는 핵심 기술 용어, 지표(Metric), 약어, 포맷 명칭 등은 억지로 한국어로 번역하지 말고 **영어 원문 그대로** 사용하세요. (예: Granularity, Outlier, Throughput, Latency, End-to-End 등) 2. **Natural Korean Phrasing**: 문장의 전체적인 구조는 자연스러운 한국어로 작성하되, 명사나 핵심 동사는 영어를 섞어 사용하는 전문적인 문체를 유지하세요. 3. **Mandatory Termi..
1. Key Terms & Definitions (핵심 용어 및 정의)PCS (Promptable Concept Segmentation): SAM 3가 새롭게 정의한 태스크로, 텍스트(예: "yellow school bus")나 Image Exemplar를 입력받아 이미지나 비디오 내에서 해당 개념(Concept)에 부합하는 모든 객체 인스턴스를 탐지, 분할(Segmentation), 추적(Tracking)하는 작업입니다.PVS (Promptable Visual Segmentation): 기존 SAM 1과 SAM 2가 수행하던 태스크로, 점(Point)이나 박스(Box) 등의 기하학적 힌트를 통해 단일 객체를 분할하는 작업입니다.Presence Head: SAM 3 아키텍처의 핵심 구성 요소로, 특정 개..
https://arxiv.org/pdf/2510.25602 내 분야는 아니지만 Gemini 3.0 Pro Preview 일단 그냥 한 번 써보고 싶어서, 그리고 요새 핫한 분야로 보여서 한 번 요약만해봄 1. Key Terms & Definitions (핵심 용어 및 정의)Quantization Granularity: 텐서 내에서 Scale Factor를 적용하는 범위를 의미합니다. 본 논문에서는 Per-tensor나 Per-channel과 같은 Coarse-grained 방식과 대비되는 Block-wise (세밀한 블록 단위) 방식을 중점적으로 다룹니다.MX Format (Microscaling): OCP(Open Compute Project)에서 제안한 데이터 포맷으로, 32개 요소(Element)마..
RZ/V2L: https://github.com/renesas-rz/rzv_drp-ai_tvm/blob/main/docs/model_list/Model_List_V2L.md rzv_drp-ai_tvm/docs/model_list/Model_List_V2L.md at main · renesas-rz/rzv_drp-ai_tvmExtension package of Apache TVM (Machine Learning Compiler) for Renesas DRP-AI accelerators powered by Edgecortix MERA(TM) Based Apache TVM version: v0.11.1 - renesas-rz/rzv_drp-ai_tvmgithub.comRZ/V2M: https://github...
회사에서 ai에이전트 개발하고 있지 않고 그냥 집에서 잠깐 시간내어 ai랑 티키타카 하여 만들어 본 예제입니다. 정답같은 예제가 아니니 참고만 하세요! 좀 마음에 안드는 부분은 인사말 건네는데 LLM이 돌아간다는 게 좀 낭비라고 생각된다. 실제로 한다면 인사말과 인사말에 대한 첫 아웃풋 토큰값들은 캐싱해놔야 비용을 절약할 수 있을 거 같다.import osimport reimport jsonimport google.generativeai as genaifrom dotenv import load_dotenvload_dotenv()# Google API 키 설정# GOOGLE_API_KEY 환경 변수에 API 키를 설정하거나, 직접 입력하세요.# 직접 입력 예: genai.configure(api_key="..
- Total
- Today
- Yesterday
- PyCharm
- ㅂ
- 순열
- 이분탐색
- FairMOT
- 백준
- 백트래킹
- C++ Deploy
- 조합
- 백준 1766
- 가장 긴 증가하는 부분 수열
- 자료구조
- 백준 11437
- 위상 정렬 알고리즘
- 문제집
- 인공지능을 위한 선형대수
- Lowest Common Ancestor
- 단축키
- LCA
- MOT
- cosine
- 파이참
- 백준 11053
| 일 | 월 | 화 | 수 | 목 | 금 | 토 |
|---|---|---|---|---|---|---|
| 1 | 2 | 3 | 4 | 5 | 6 | 7 |
| 8 | 9 | 10 | 11 | 12 | 13 | 14 |
| 15 | 16 | 17 | 18 | 19 | 20 | 21 |
| 22 | 23 | 24 | 25 | 26 | 27 | 28 |
| 29 | 30 | 31 |
