Kylin Page

A fool who dreams.

AlpaServe Distributed ML Serving

Statistical Multiplexing with Model Parallelism for Deep Learning Serving

[TOC] This work demonstrate that model parallelism can be additionally used for the statistical multiplexing of multiple devices when serving multiple models, even when a single model can fit into...

Alpa Distributed ML Compiler

Automating Inter- and Intra-Operator Parallelism for Distributed Deep Learning

[TOC] Architecture is all you need! Framework 主要的idea:: 结合 intra-op 和 inter-op 两种并行策略,进行比较好的计算图划分、设备分配策略 什么是 intra-op 和 inter-op ? Intra-op: 对数据对划分、对算子的划分 inter-op: 对计算图的划...

Python3 Cookbook

fast introduction to python3

[TOC] 本文总结了三个教程,原教程链接地址: Python tutorial:http://www.pythondoc.com/pythontutorial3/index.html Python for you and me:http://pymbook.readthedocs.io/en/latest/ The Python Standard L...

Lecture Function as a Service

Notes for Lecture from Boris Grot

Poster intro serverless a collection of functions : dont have states Challenge: communication bottleneck we dont have severless framework for academic research vHive: https://github.com/vh...

novel ideas for MLLM research

comprehensive survey for MLLM research

[TOC] idea: rethinking 传统的 image caption 的问题(从前研究没法解决的LLM可以解决的) caption组合泛化外部知识结合 以人为中心的 Video Caption Dense Video Caption 的动态 语言多样性的 Video Caption 细粒度的 Caption 新闻文本的caption/短视频评论的...

MacawLLM

MULTI-MODAL LANGUAGE MODELING WITH IMAGE, AUDIO, VIDEO

[TOC] Abstract 多模态LLM:crossing visual/audio/text three main components: modality module for encoding multi-modal data cognitive module for har- nessing pretrained LLMs ...

Jupyter Notebook 中的 Magic Commands

魔术命令 Cheatsheet

[TOC] 单命令模式 %run 运行外部Python脚本或Jupyter Notebook文件 %run my_script.py %time %timeit 测量代码块的执行时间 %time sum(range(1000)) %timeit sum(range(1000)) %load 加载外部Python脚本或文本文件的内容到一个代码单元格。 %load my_sc...

Notes for M3IT

A LargeScale Dataset towards MultiModal Instruction Tuning

[TOC] Abstract 给了个VLM进行 Instruction Tuning 的数据集:https://huggingface.co/datasets/MMInstruction/M3IT 在M3IT上训了一个Ying-VLM,据说多模态能力很强 Experiment Setting 8 NVIDIA 80GB A100 GPUs. It took about 10 day...

Learning Notes in Kwai

Learn in Kwai

[TOC] Notes 2>&1 https://segmentfault.com/a/1190000040086046 nohub之后怎么知道PID 看stdout(or重定向之后的输出文件) ps aux | grep "your_command" 输出上一个执行进程的PID echo $! 已知PID查看进程详细信息 ...

Notes for BLIP2

VQA

[TOC] Notes catastrophic forgetting Catastrophic forgetting(灾难性遗忘)是指在连续学习或迁移学习中,当一个机器学习模型学习新任务时,它会忘记之前学到的旧任务的能力的现象。当一个模型被训练来解决多个任务时,通常会发生这种情况。 zero-shot instructed image-to- text generat...