栗子研-优快云博客

原创【flash-attn】Linux安装踩坑记录（集群安装版）

本文记录了在学校服务器上安装flash-attn时遇到的典型问题及解决方案。由于服务器登录节点无GPU而计算节点无网络，导致直接pip安装失败。尝试下载预编译轮子时又遇到GLIBC版本不兼容问题（2.28 vs 2.32）。随后尝试源码编译，但发现GPU架构不匹配（A800最高支持8.0），需修改setup.py中的CUDA_ARCHS设置。最终采用降版本策略，成功找到兼容的wheel文件完成安装。整个过程凸显了在受限环境中安装CUDA依赖的复杂性，建议优先寻找版本兼容的预编译包而非耗时编译。

2025-12-07 13:30:12 1497 1

原创【Chroma】Cannot return the results in a contigious 2D array. Probably ef or M is too small

摘要：使用LangChain(0.3.20)和ChromaDB(0.6.3)时，在构建向量数据库进行相似性搜索时出现RuntimeError，提示ef或M参数过小导致无法返回2D数组。解决方法是在创建Chroma向量库时通过collection_metadata参数调整HNSW配置，将M值从默认的16提高到48，搜索ef从100提升至200，构建ef从100增至400。文中对比了默认参数与优化后的配置，解决了搜索时的数组连续性错误问题。

2025-05-30 17:34:06 318

原创【阅读笔记】Large Language Models Know What is Key Visual Entity: An LLM-assisted Multimodal Retrieval for

题目：Large Language Models Know What is Key Visual Entity: An LLM-assisted Multimodal Retrieval for VQA会议：EMNLP 2024 (CCF-B会）作者：Pu Jian（中科大），Donglei Yu（中科大），Jiajun Zhang（中科大）

2025-01-08 12:04:35 1253 1

原创【阅读笔记】Improving Zero-shot Visual Question Answering via LargeLanguage Models with Reasoning Question

题目：Improving Zero-shot Visual Question Answering via Large Language Models with Reasoning Question Prompts会议：ACMMM（ACM International Conference on Multimedia） 2023 （CCF-A会）

2024-12-22 22:20:43 1113

原创【阅读笔记】Prompting large language model with context and pre-answer forknowledge-based VQA

题目：Prompting large language model with context and pre-answer for knowledge-based VQA期刊：PR 2024（Pattern Recogniztion）（CCF-B类）作者：Zhongjian Hu, Peng Yang, Yuanshuang Jiang, Zijian Bai （东南大学）

2024-12-22 12:02:45 886

原创【阅读笔记】Zero-shot Visual Question Answering using Knowledge Graph

题目：Zero-shot Visual Question Answering using Knowledge Graph期刊：ISWC 2021（B类会议）作者：陈卓(浙江大学)、陈矫彦(牛津大学)、耿玉霞(浙江大学)、Jeff Z. Pan(爱丁堡大学)、苑宗港(华为)、陈华钧(浙江大学)

2024-11-04 20:24:48 773

weixin_60074434的博客

原创【flash-attn】Linux安装踩坑记录（集群安装版）

原创【Chroma】Cannot return the results in a contigious 2D array. Probably ef or M is too small

原创【阅读笔记】Large Language Models Know What is Key Visual Entity: An LLM-assisted Multimodal Retrieval for

原创【阅读笔记】Improving Zero-shot Visual Question Answering via LargeLanguage Models with Reasoning Question

原创【阅读笔记】Prompting large language model with context and pre-answer forknowledge-based VQA

原创【阅读笔记】Zero-shot Visual Question Answering using Knowledge Graph

空空如也

空空如也

原创 【flash-attn】Linux安装踩坑记录（集群安装版）

原创 【Chroma】Cannot return the results in a contigious 2D array. Probably ef or M is too small

原创 【阅读笔记】Large Language Models Know What is Key Visual Entity: An LLM-assisted Multimodal Retrieval for

原创 【阅读笔记】Improving Zero-shot Visual Question Answering via LargeLanguage Models with Reasoning Question

原创 【阅读笔记】Prompting large language model with context and pre-answer forknowledge-based VQA

原创 【阅读笔记】Zero-shot Visual Question Answering using Knowledge Graph

空空如也

空空如也

原创【flash-attn】Linux安装踩坑记录（集群安装版）

原创【Chroma】Cannot return the results in a contigious 2D array. Probably ef or M is too small

原创【阅读笔记】Large Language Models Know What is Key Visual Entity: An LLM-assisted Multimodal Retrieval for

原创【阅读笔记】Improving Zero-shot Visual Question Answering via LargeLanguage Models with Reasoning Question

原创【阅读笔记】Prompting large language model with context and pre-answer forknowledge-based VQA

原创【阅读笔记】Zero-shot Visual Question Answering using Knowledge Graph