2025. 10. 10. Friday다음주 랩미팅 준비: Code Review · 구현 📌 참고자료 - UPMEM SDK: https://sdk.upmem.com/stable/030_DPURuntimeService_Tasklets.html- UPMEM Naive-GEMM: https://github.com/hhessammheidary/UPMEM-GEMM 📌 TODOLIST- UPMEM Checksum example Code Review- PIM Embedding Lookup(Python Wrapper) - MHA Implementation

ZeroMQ Patterns Request-reply, which connects a set of clients to a set of services. This is a remote procedure call and task distribution pattern (REQ, REP, DEALER, ROUTER). 요청하면 응답을 보내준다. 내가 요청하면 가운데 애가 받아서, 일꾼이 3명 있는데 어떤 애한테 보낼까? 니가 가져가라 등등 적절히 분배하는 과정도 이미 구현됨 이 요청에 대해서는 니가 처리했음 좋겠다 -> 설정하는 것도 구현됨 Pub-sub, which connects a set of publishers to a set of subscribers. This is a data distribution..

# Dockerfile WebSocket https://leimao.github.io/blog/Boost-Docker/ Boost C++ Library Docker ContainerPortable and Convenient C++ Libraryleimao.github.io                         👩‍💻Docker desktop을 통해 쉘을 띄워서 클라이언트를 열어도 되고, (서버와 클라이언트를 하나의 이미지 파일로 빌드한 경우) 서로 다른 이미지로 분리하여, 통신을 시킬 수도 있습니다!     🐳 제출한 Docker hub link https://hub.docker.com/repository/docker/sohyeonkimdev/websocket_docker/general Doc..

1. ETH Advanced * Trillema – 1. Security - The ability of the blockchain system to operate as expected, defend itself from attacks, bugs, and other unforeseen issues : 보안 2. Scalability - The ability for a blockchain system to handle an increasingly growing amount of transactions : 확장성 3. Decentralization - Creating a blockchain system that does not rely on a central point of control - 이더리..

2025. 12. 10. Wednesday

2025. 12. 03. Wednesday * UPMEM PIM Programming- dpu.c : tasklet_id = me(), 한 DPU 내부의 tasklet 구분 - host.c : 서로 다른 DPU 인식 -> 구현을 여기서 수정해야 한다. - 이때 DPU_FOREACH Iterator를 활용한다. - As a consequence, the host API works on sets of DPUs, which may contain multiple DPU ranks.- The provided C macro DPU_RANK_FOREACH and DPU_FOREACH iterate over the ranks and DPUs respectively of a set. # UPMEM-GEM..

2025. 11. 29 Saturday 한 달 내내 풀지 못했던 에러를 해결했다.!너무너무너무 기뻤다. .·⋆(⌒_⌒)⋆·. 오랜만에 느껴본 코딩(디버깅)의 맛 .막막하고 어려운 문제도 step by step으로생각해보면, 찬찬히 해결할 수 있다 ! * dpu-diag- 총 2546개 DPU, 350Mhz를 확인할 수 있다. - perfcounter_get 로 가져온 cycle을 ms로 표현할 때, 정확한 hz가 필요하다. * Cycles 선형 증가 해결 - perfcounter_config(COUNT_CYCLES, true); - 초기화 구문 추가로, head 별 cycle 에러 해결 * DPU 내부 float 연산 제거 - 구조체 정의 변경, PIM 성능 향상 + TILE_ROWS 최적값..

2025. 10. 29. Wednesday

1. history of blockchain - The Genesis Block of Bitcoin - Genesis block = 블록체인 최초의 0번 블록 : 사토시나카모토, 중앙은행, 정부의 무능으로부터 자유로운 탈중앙화 화폐 - “The Times 03/Jan/2009 Chancellor on brink of second bailout for banks.” 2. Blockchain Technologies * 합의 모델 : 분산 컴퓨팅에서의 고전적 문제- faulty = 나쁜놈, 이상한 놈이 섞인 환경 -> 어떻게 합의(같은 값)에 도달? - 비잔틴 장군의 문제 = 2/3 이상의 장군들 충성 -> N >= 3f + 1 - PBFT (Practical Byzan..

2025. 10. 29. Wednesday 📌 TODOLIST - quantization 적용 더 알아보기 (float -> int) - dpu 2nd tasklet error debugging - embedding 차원 조절하기 (tasklet >= 16)

2025. 10. 11. SaturdayUPMEM Official Example Review

2025. 10. 10. Friday다음주 랩미팅 준비: Code Review · 구현 📌 참고자료 - UPMEM SDK: https://sdk.upmem.com/stable/030_DPURuntimeService_Tasklets.html- UPMEM Naive-GEMM: https://github.com/hhessammheidary/UPMEM-GEMM 📌 TODOLIST- UPMEM Checksum example Code Review- PIM Embedding Lookup(Python Wrapper) - MHA Implementation

2025. 10. 01. 수요일

2025. 09. 30. 화요일

UPMEM Hello World! Examplehttps://sdk.upmem.com/stable/02_HelloWorld.html Hello World! Example — UPMEM DPU SDK 2025.1.0 Documentation© Copyright 2015-2024, UPMEM SAS - All rights reserved.sdk.upmem.com 0. UPMEM SDK 설치하기 https://sdk.upmem.com UPMEM DPU SDKUPMEM SDK The Software Development Kit for programming and using the DPU provided by the UPMEM Acceleration platform.sdk.upmem.com tar -..

PAPI: Exploiting Dynamic Parallelism in Large Language Model Decoding with a Processing-In-Memory-Enabled Computing Systemhttps://arxiv.org/abs/2502.15470 PAPI: Exploiting Dynamic Parallelism in Large Language Model Decoding with a Processing-In-Memory-Enabled Computing SystemLarge language models (LLMs) are widely used for natural language understanding and text generation. An LLM model relies ..

Pimba: A Processing-in-Memory Acceleration for Post-Transformer Large Language Model Servinghttps://github.com/casys-kaist/pimba GitHub - casys-kaist/pimba: Official code repository for "Pimba: A Processing-in-Memory Acceleration for Post-Transformer LargeOfficial code repository for "Pimba: A Processing-in-Memory Acceleration for Post-Transformer Large Language Model Serving [MICRO'25]" - casys..

Accelerating LLMs using an Efficient GEMM Library and Target-Aware Optimizations on Real-World PIM Devices * TVM = deep learning compiler frameworkApache TVM is a machine learning compilation framework, following the principle of Python-first development and universal deployment. It takes in pre-trained machine learning models, compiles and generates deployable modules that can be embedded and..

2025. 06. 20 금요일

2025. 06. 10. 화요일 📌 TODOLIST- FLIP PROB 0.00001 수준으로 줄이기- 동일한 rate에서 100회 반복 후 통계 - 값이 보존 or 변경 여부로 그래프 - random seed 변경하도록 구현

2025. 06. 05. 목요일

2025. 05. 27. Tuesday

2025. 05. 20. 화요일 Paper: https://open.library.ubc.ca/soa/cIRcle/collections/ubctheses/24/items/1.0435518 Offloading embedding lookups to processing-in-memory for deep learning recommender modelsRecommender systems are an essential part of many industries and businesses. Generating accurate recommendations is critical for user engagement and business revenue. Currently, deep learning recomme..