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
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Serial orderings, parallel layerings and spatial partitions for D8 flow networks, computed once from the flow-direction grid and reused by any drainage-order computation. Python reference implementation.
翻译准备中
取自仓库 README
正在加载完整 README。 以下为已保存的摘录。
Orderings, layerings and partitions for D8 flow networks, in Python.
A D8 flow-direction grid fixes which cells must be done before which: a cell
needs its upstream neighbours first, or, for some kernels, its downstream one.
FlowTopo works that order out once, from the grid alone, and saves it.
Anything that walks the network in drainage order (upstream drainage area,
flow length, Strahler stream order, or the upstream-to-downstream step of a
routing scheme)
then reads the saved order instead of working it out again.查看完整 README ↗FlowTopo builds reusable orderings, parallel layerings and spatial partitions from a D8 flow-direction grid.
Three serial orderings visit every cell once while respecting the flow dependencies.
git clone https://github.com/hellolulujiang/flowtopo.git
cd flowtopo
pip install -e ".[speed]"
Python 3.10 or newer. The speed extra enables threaded kernels.
import flowtopo
topo = flowtopo.FlowTopo.from_raster("data/dir_example.tif")
upa = topo.upstream_area(ordering="dfs") # km²
upa_grid = topo.to_2d(upa)
The bundled MERIT Hydro basin has 93,432 valid cells. Input requirements.
Three parallel layerings. Layers run in order; the cells of one layer run at the same time.
Three propagation manners. The structure decides when a cell runs; the manner decides who writes the result.
FullTopo v1.1 provides saved structures for the 90 MERIT Hydro regions of FullDivide v1.1. Contents and use.
User guide · Methods and provenance · Notebook · Video · Animations · FullHydro tools
Tests ·
Cite this repository with the release used and downloaded products by version. Paper: MERIT-FlowTopo v1.0: a reusable computational foundation for hyperresolution hydrology on the global 90 m drainage network (Jiang et al., submitted to Geoscientific Model Development). Archived paper code.
MERIT Hydro (Yamazaki et al., 2019) supplies the example grid. Foundational conventions follow pyflwdir (D. Eilander, Deltares; MIT); see provenance.
MIT; see LICENSE. The bundled MERIT Hydro excerpt retains its terms, CC BY-NC 4.0 or ODbL 1.0; see DATA_NOTICE.md.