neuron-explorer
Steering-first neuron explorer for Qwen3.5-4B: push every MLP neuron, keep every rollout, browse the results
翻译准备中
尚未检测到更新。GitHub 发布版本后会显示在这里。
个人 GitHub 账号下公开仓库的所有者可以登录并验证所有权,自行管理介绍,无需管理员批准。
登录并生成验证文件,在 30 分钟内将其提交到此仓库的默认分支。其他分支或分叉仓库中的文件无效。然后在下方检查文件,即可无需管理员批准直接管理项目。请保留文件以维持管理权限。
如需登录,将前往登录页面。登录后请回到此处,再次点击所选的验证按钮。
在项目文件夹中运行 /nomorevibe verify,即可验证所有权并自行管理介绍。
一分钟完成验证正在加载完整 README。 以下为已保存的摘录。
README
======
docx4j is an open source (Apache v2) library for creating, editing, and saving OpenXML "packages", including docx, pptx, and xslx.
It uses JAXB to create the Java representation.查看完整 README ↗docx4j is an open source (Apache v2) library for creating, editing, and saving OpenXML "packages", including docx, pptx, and xslx.
It uses JAXB to create the Java representation.
Evaluating your options for working with Word documents?
pip install docx4j, no JVM), and when to call Java docx4j from Python insteaddocx4j 17.0.0 targets Java 11. It is compatible with Java 11, 17, 21, and we expect Java 25.
It uses Jakarta XML Binding API 4.0 (as did docx4j v11.5.x)
Version numbering jumped from 11.5.14 to 17.0.0 to flag three significant changes:
docx4j-generated-objectsdeepCopyFast method which is used unless configured otherwise to copy objects (ie instead of marshal/unmarshal)getEGBlockLevelElts, getParagraphContent, getRunContent have been replaced with getContent;
ArrayListWml, ArrayListDml, ArrayListVml, and ArrayListMce have been replaced by a single ArrayListDocx4j class;
org.jvnet.jaxb2_commons.ppp.Child is now org.jvnet.jaxb.lang.ChildTo use it, add the dep corresponding to the JAXB implementation you wish to use
docx4j v11.4.5 uses Jakarta XML Binding API 3.0, as opposed to JAXB 2.x used in earlier versions (which import javax.xml.bind.*). Since this release uses jakarta.xml.bind, rather than javax.xml.bind, if you have existing code which imports javax.xml.bind, you'll need to search/replace across your code base, replacing javax.xml.bind with jakarta.xml.bind. You'll also need to replace your JAXB jars (which Maven will do for you automatically; otherwise get them from the relevant zip file).
Being a JPMS modularised release, the jars also contain module-info.class entries.
This is docx4j for Java 8. Although in principle it would compile and run under Java 6, some of its dependencies are Java 8 only. So to run it under Java 6, you'd need to use the same version of the deps which docx4j 6.x uses.
docx4j v8 is a multi-module Maven project.
To use docx4j v8, add the dep corresponding to the JAXB implementation you wish to use. You should use one and only one of docx4j-JAXB-*
Get it from GitHub, at https://github.com/plutext/docx4j
mvn clean
mvn install -Dgpg.skip=true
For more details, see http://www.docx4java.org/blog/2015/06/docx4j-from-github-in-eclipse-5-years-on/
http://www.docx4java.org/downloads.html
See the Getting Started guide: docs/Docx4j_GettingStarted.md (also there as docx, HTML and PDF — the docx is canonical; other formats are generated from it by docx4j itself)
and the Cheat Sheet: http://www.docx4java.org/blog/2013/05/docx4j-in-a-single-page/
And see the sample code in the docx4j-samples-* modules, eg
docx4j-samples-docx4j
You'll probably want the Helper AddIn to generate code: http://www.docx4java.org/blog/2016/05/docx4j-helper-word-addin-new-version-v3-3-0/
LLMs and coding agents are good at writing docx4j code, and you should feel
free to use them — we recommend Claude. Two things
help: docx4j's documentation is available in LLM-friendly form at
https://www.docx4java.org/llms.txt, and the docx4j-samples-* modules give an
agent working code to pattern-match against. The classic failure modes to
watch for: javax.xml.bind imports (current docx4j uses jakarta.xml.bind),
and invented APIs for merging documents (merging is a commercial extension —
see https://www.plutext.com).
The source tree is set up for AI-assisted development too: CLAUDE.md orients a coding agent (build commands, module map, architecture), and docs/developer/change-requests/ records the design history behind non-trivial changes. Much of docx4j 17.x was built this way.
Contributions — including AI-assisted ones — are welcome. Please do raise issues (a small docx reproducing the problem is worth a thousand words) and pull requests; see CONTRIBUTING.md for the ground rules.
To talk to a human, use GitHub discussions.
Your LLM is a good first resort (see "Using an LLM with docx4j" above).
To talk to a human, use GitHub discussions; there is also the forum at http://www.docx4java.org/forums and StackOverflow (tag 'docx4j'). Please post to one, not several.
Contributions are welcome; please see CONTRIBUTING.md
before opening a pull request. In short: base your PR on the current
VERSION_x_y_z development branch, sign off each commit (DCO,
git commit -s), and disclose any AI assistance with a commit trailer.
docx4j is published under the Apache License version 2.0. For the license text, please see the following files in the legals directory: