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
每日模型推荐,针对 GitHub Copilot、OpenCode Go、OpenCode Zen 和 ClinePass,按任务提供——以小餐馆风格呈现。
AI 翻译
Daily model recommendations for GitHub Copilot, OpenCode Go, OpenCode Zen and ClinePass, per task — served bistro-style.
开发工具TypeScript · HTMLGitHub @KuzniAI在线运营者未确认
试用过了吗? 点一下就好,无需登录。
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。
一分钟完成验证选择最适合您的洋葱卖家。
AI 翻译
Choose a best onion seller for you
取自仓库 README
正在加载完整 README。 以下为已保存的摘录。
Choose a best onion seller for you.
1. GitHub Copilot https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing
2. OpenCode Go https://opencode.ai/docs/go/
3. OpenCode Zen https://opencode.ai/docs/zen/
4. ClinePass https://docs.cline.bot/getting-started/clinepass
config/model-mapping.json links each provider model to its Artificial Analysis slug. Every run updates it:查看完整 README ↗Choose a best onion seller for you.
config/model-mapping.json links each provider model to its Artificial Analysis slug. Every run updates it:
(Reasoning) are ignored; the base, unsuffixed AA slug is used).aaSlug: null are retried each run, and entries whose slug disappeared from AA data are re-matched.aaSlug still exists in AA data are never touched, so you can override a match by editing aaSlug by hand.Pricing parsers (src/<provider>/parser.ts) are written and repaired by a Pi coding agent through the Pi SDK (@earendil-works/pi-coding-agent), not by hand. When a parser throws, src/pricing/repair-parser.ts hands the downloaded page to Pi:
data/<provider>/source.html. Pi is told to rewrite only that provider's parser.ts against it, following the row contract in the provider's source.ts, and to run lint and format.git diff src/) and commit it.Repair is off by default and never runs in CI:
PARSER_REPAIR=1 in .env turns it on. It is ignored whenever CI or GITHUB_ACTIONS is set, so in the daily workflow a broken parser just fails the run.PI_REPAIR_PROVIDER and PI_REPAIR_MODEL pick the Pi model (opencode-go / glm-5.3-flash in .env.example).npx pi and /login, or set the provider's API key env var (OPENCODE_API_KEY for opencode-go).OpenCode Zen and ClinePass start with stub parsers that throw. Run npm start locally once with PARSER_REPAIR=1 to have Pi generate them. The daily workflow fails until those generated parsers are committed.
.github/workflows/daily.yml runs the pipeline every day at 05:17 UTC (or on demand from the Actions tab):
config/model-mapping.json are committed straight to main.data branch: latest/ holds the current recommendations.json, provider pricing, and the raw Artificial Analysis data; history/YYYY-MM-DD/ keeps one snapshot per day.site/index.html is published with the latest data to https://kuzniai.github.io/onion-sellers/, including the Artificial Analysis data's fetch date so a rate-limited run is visible.One-time setup:
AA_API_KEY repository secret (Settings → Secrets and variables → Actions).The workflow declares its own contents: write and pages: write permissions, so the repository's default workflow permissions can stay read-only. Pushing mapping changes requires main to accept pushes from github-actions[bot].
GitHub disables scheduled workflows in public repositories after 60 days without activity; re-enable it in the Actions tab if that happens.