重點導讀
自己微調 AI 模型,常要準備資料、設定訓練流程,還得另外檢查結果。
- Overmind 推出一種用手機和 Grok 聊天就能操作的方式:你上傳自己標好類別的例子,再告訴 Grok 想讓模型學什麼。
- Grok 會透過 Overmind 的 MCP 工具讀取資料並開始訓練,Overmind 則用自家的 GPU 微調 Qwen3.5-0.8B。訓練後,系統會用沒參加訓練的測試資料比較原模型和微調後模型的分數,再交給 Grok 顯示,讓你檢視前後差異。
AI 伴隨問答
針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:
Rohan Paul@rohanpaul_ai
·3 天前
Overmind just launched a way to fine-tune your own AI model by chatting with Grok on your phone.
You upload your own labeled examples, tell Grok what to learn, and it trains a small model, then scores it before and after on test cases it never saw.
- Grok connects to @OvermindLab’s MCP server, which gives it a set of tools: read your dataset, start training, run evaluations.
- When you describe the task in chat, Grok turns it into tool calls, and Overmind’s own GPUs fine-tune Qwen3.5-0.8B on the rows marked “train.” Overmind then runs both the original and the fine-tuned model on the rows marked “test” and sends both scores back to Grok, which shows them to you.
推文來源 · X (Twitter)在 X 參與討論
來源: Rohan Paul (x.com)
在 X 查看推文