AI Paper Atlas

Research map / Optimization / Optimizers

Mechanic: A Learning Rate Tuner

Ashok Cutkosky, Aaron Defazio, Harsh Mehta

NeurIPS 2023 · Optimization › Optimizers · approach: Memory-efficient LLM optimizers · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
pre-train with Mechanic instead of tuned AdamWBERT-B · MLMMLM accuracy: 71.5 -> 71.7Improved
pre-train with Mechanic instead of tuned LionBERT-B · MLMMLM accuracy: 71.8 -> 72.0Improved
pre-train with Mechanic instead of tuned AdamWBERT-L · MLMMLM accuracy: 75.4 -> 75.3No clear effect
pre-train with Mechanic instead of tuned LionBERT-L · MLMMLM accuracy: 75.7 -> 75.5No clear effect
pre-train with Mechanic instead of tuned Lion, then finetune with LionBERT-B · GLUE QNLIaccuracy: 86.8 -> 89.0Improved
pre-train with Mechanic instead of tuned optimizer on JFT-300MViT-B/16 with Lion · JFT-300Mpre-training accuracy: 47.0 -> 49.6Improved
pre-train with Mechanic instead of tuned optimizer on JFT-300MViT-L/16 with Lion · JFT-300Mpre-training accuracy: 52.0 -> 55.4Improved
pre-train with Mechanic instead of tuned AdamW on JFT-300MViT-L/16 · JFT-300Mpre-training accuracy: 54.4 -> 54.4No clear effect
use Mechanic on small dataset from scratchResNet-18 Momentum · CIFAR-100accuracy: 77.6 -> 75.3Hurt
raise weight-decay-like term lambda from 0.01 to 0.1 (ablation)Mechanic ResNet-18 · CIFAR-100accuracy: 75.3 -> 76.6Improved
raise lambda from 0.01 to 0.1 (ablation)Mechanic ResNet-18 · CIFAR-10accuracy: 94.1 -> 95.3Improved
use Mechanic instead of tuned MomentumWRN-40-10 · CIFAR-100accuracy: 79.9 -> 79.6No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy