Skip to content

Releases

v1.4.1

Choose a tag to compare

@Honguan Honguan released this 16 Oct 16:36
14a0e44

依第一次真的 GPU 實測(RTX 4060 Laptop 8 GB、Windows 11,10 項中 9 項通過)做的修正。

修正

  • eval gpu:VRAM 峰值改成「最高減最低讀數」;量不到 VRAM 變化時只提出警告,不再判為失敗(實際裝置以引擎記錄為準)。這次唯一的失敗就是這個誤判。
  • eval gpu 的 Whisper 檢查改用真的英文語音(LibriSpeech 片段,需要 pyarrow),要求 WER ≤ 0.5;embedding 檢查改用一對一的問答,要求 recall@1 ≥ 0.8。
  • llm-finetune 0.1.2、llm-preference 0.1.1:最低 VRAM 宣告從 8 GB 改成 6 GB(0.5B QLoRA 實測 1.4–1.6 GB),6 GB 的顯示卡也能用;gradient checkpointing 改用 use_reentrant=False,不再出現警告。
  • text-classification 0.1.1:記錄 loss 改用 loss.item()。

實測結果:https://honguan.github.io/autoforge/GPU/

升級並重新驗證

pip install -U autoforge-trainer pyarrow
autoforge eval gpu

完整變更紀錄:https://github.com/Honguan/autoforge/blob/main/CHANGELOG.md

v1.4.0

Choose a tag to compare

@Honguan Honguan released this 16 Oct 08:48
874bad6

新功能

  • 在 GPU 上驗證方法包:autoforge eval gpu [--full]。10 個 GPU 方法包各用一小份容易學的資料,在 CUDA 上真的訓練(QLoRA 4-bit、混合精度、預訓練模型),一邊用 NVML 量 VRAM 峰值,再註冊並預測一次。報告列出每一項的結果、時間、VRAM 峰值和方法包宣告的最低 VRAM。說明:https://honguan.github.io/autoforge/GPU/

修正

  • image-classification 0.1.1、llm-finetune 0.1.1:記錄 train_loss 時改用 loss.item(),不再出現 PyTorch 警告。
  • autoforge index add 更新方法包時,沒給 --tag 會保留原本的搜尋標籤;索引裡每個方法包都補上了標籤,包括中文關鍵字(例如「推薦」「語音辨識」)。

在你的 GPU 上試試

pip install -U "autoforge-trainer[all]" bitsandbytes
autoforge eval gpu

完整變更紀錄:https://github.com/Honguan/autoforge/blob/main/CHANGELOG.md

v1.3.0

Choose a tag to compare

@Honguan Honguan released this 16 Oct 07:56
dfd5109

新功能

  • 評估 AI 選方法的準確度:autoforge eval agent --brain <大腦>。13 個案例,每個是一句需求加一份當場產生的小資料,涵蓋表格、文字、圖片、物件偵測、時間序列、異常偵測、推薦、文字向量、LLM 微調和敏感資料。AI 要開始訓練時會被攔下並記錄計畫,不會真的訓練;結果分成 pass / partial / fail,報告存在 ~/.autoforge/evals/。

改進

  • 資料分析:資料夾會列出最上層的檔案(例如 annotations.csv),JSONL 會列出欄位名稱和筆數(不含內容),讓 AI 分得出物件偵測資料、對話資料和配對資料。

安裝與試用

pip install -U autoforge-trainer
export ANTHROPIC_API_KEY=...
autoforge eval agent --brain anthropic:claude-opus-5-5 -c churn -c labelled-fraud # 先跑兩題
autoforge eval agent --brain anthropic:claude-opus-5-5 # 全部 13 題

完整變更紀錄:https://github.com/Honguan/autoforge/blob/main/CHANGELOG.md

v1.2.0

Choose a tag to compare

@Honguan Honguan released this 16 Oct 07:10
77a4198

新功能

  • 網頁介面的 AI 助手:在 autoforge gui 裡和內建 AI 代理對話。可以選 AI 大腦(Claude、OpenAI、本機 Ollama),用一句話說需求;AI 開始訓練、重新訓練或自動調參前會在頁面上請你同意,不同意時可以寫意見讓它修改;訓練完可以接著問。AI 用自己的身分操作,隱私規則和指令列相同。
  • AI 代理改成可以多輪對話(AgentSession),autoforge train 的行為不變。

修正

  • 修正幾個還寫著舊套件名稱 autoforge[...] 的安裝提示。

安裝

pip install -U autoforge-trainer
autoforge gui # 上方選單「AI 助手」

完整變更紀錄:https://github.com/Honguan/autoforge/blob/main/CHANGELOG.md

v1.1.0

Choose a tag to compare

@Honguan Honguan released this 16 Oct 06:30
ff2fc19
Add object-detection and image-segmentation packs; end-to-end tests f…

v1.0.2

Choose a tag to compare

@Honguan Honguan released this 16 Oct 04:29
e708f50
Rename PyPI distribution to autoforge-trainer (autoforge-ml was taken…