AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
翻譯準備中
建構 RL 與 SFT 資料集很困難。whileai 幫您完成:模擬、以驗證器或人工檢查的評審評分、以區間測量 pass@1、保留有訊號的列,使用 SFT 與 RL 訓練。The defaul
AI 翻譯
Building RL and SFT datasets is hard. whileai does it: simulate, grade with a verifier or a checked judge, measure pass@1 with an interval, keep the rows with signal, train with SFT and RL. The defaul
開發工具Python · PLpgSQLGitHub @whilehq上線中營運者未確認
試用過了嗎? 點一下就好,不需要登入。
尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify,即可驗證所有權並自行管理介紹。
一分鐘完成驗證Scientific RL and SFT post-training for AI agents: build evals that can fail, simulate situations, judge checked against people, pass@1 with a 95% interval, train with GRPO/SFT/DPO, prove every gain on a held-out set. uv add whileai
翻譯準備中
取自儲存庫 README
正在載入完整 README。 以下為已儲存的摘錄。
• Simulate. Give it your agent's tools and prompt. It writes the
situations the agent will meet and runs them against a fake world that
fails on purpose.
• Grade. Your judge or a verifier scores every rollout. The judge is
checked against people before its scores count.
• Measure. One run proves nothing. Ask whether a change is real or
noise before you ship it or train on it.
• Select. Keep the rows that carry signal: the 20 to 80% band for RL,
the best completion for SFT, nothing that leaks into your eval set.
• Train.檢視完整 README ↗目前無法載入完整 README。請透過原文連結閱讀。 以下為已儲存的摘錄。
• Simulate. Give it your agent's tools and prompt. It writes the
situations the agent will meet and runs them against a fake world that
fails on purpose.
• Grade. Your judge or a verifier scores every rollout. The judge is
checked against people before its scores count.
• Measure. One run proves nothing. Ask whether a change is real or
noise before you ship it or train on it.
• Select. Keep the rows that carry signal: the 20 to 80% band for RL,
the best completion for SFT, nothing that leaks into your eval set.
• Train.檢視完整 README ↗