跳到正文
𝕏 elvis(x.com)·· 3 天前品質評分52
重點導讀

模型量化時,適合的設定會隨模型和精度改變,專家往往得手動重做。

  • MIT 和 NVIDIA 的論文提出 QDA,讓 AI 助理一輪輪組合並測試量化方案。
  • 它先用幾分鐘比較候選模型和原模型在推理過程中的輸出差異,再對有希望的方案進行較長測試。論文表示,QDA在幾種設定下都勝過最好的公開方案。
  • 在 320B MoE GLM-5.3-Flash 的 NVFP4 W4A4 測試中,保留了 FP8 準確度的 98.3%,高於最佳廠商檢查點的 96.9%。

AI 伴隨問答

自備 Key 深度分析

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:

elvis
elvis@omarsar0
·3 天前
𝕏

Great paper from MIT and NVIDIA.

I think this is an important step toward personalized, accessible RSI.

The best recipe changes with every model and precision, so experts usually redesign it by hand. QDA hands that loop to an agent that builds and tests candidates over many rounds.

They show that QDA beats the strongest public recipes by 0.9, 1.4 and 2.2 points at W4A16, W4A8 and W3A16, using 64 GPU-hours per search.

Evaluation speed is what makes this work. Scoring one quantized reasoning model normally takes tens of GPU-hours. QDA scores each candidate in minutes by measuring its output KL divergence from the original model on reasoning traces the agent never sees. Promising candidates then get long generations to catch repetition loops.

A shared library of public PTQ methods lets the agent combine them without porting each codebase by hand.

On the 320B MoE GLM-5.3-Flash at NVFP4 W4A4, QDA recovered 98.3% of FP8 accuracy, against 96.9% for the best vendor checkpoint. Every checkpoint runs on existing serving kernels with no inference overhead.

https://t.co/1RFcsXN7jV

Chat with Paper: https://t.co/Kq1tKeqho6

推文來源 · X (Twitter)在 X 參與討論
來源: elvis (x.com)
在 X 查看推文

讀者討論

理性交流 · 即時討論