From 8fc9ba2e9e7a8d7c72cdaeb36d61ea485d1d5819 Mon Sep 17 00:00:00 2001 From: Thien Tran Date: Sat, 6 Jul 2024 18:43:40 +0800 Subject: [PATCH] update --- torchao/prototype/low_bit_optim/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/torchao/prototype/low_bit_optim/README.md b/torchao/prototype/low_bit_optim/README.md index 7a108959f7..5c1d631d2c 100644 --- a/torchao/prototype/low_bit_optim/README.md +++ b/torchao/prototype/low_bit_optim/README.md @@ -31,7 +31,7 @@ NOTE: Benchmark script for fine-tuning a [timm](https://github.com/huggingface/pytorch-image-models) model on [resisc45](https://huggingface.co/datasets/timm/resisc45) dataset is available at [benchmarks/benchmark_low_bit_adam.py](../../../benchmarks/benchmark_low_bit_adam.py). -Results for fine-tuning ViT-H (630M params) with BF16 AMP, batch size 8, on 4070Ti SUPER, with fixed random seed: +Results for fine-tuning ViT-H (630M params) with BF16 AMP for 2 epochs, batch size 8, on 4070Ti SUPER, with fixed random seed: Adam impl | max memory (GB) | time taken for 2nd epoch | accuracy ---------------|-----------------|--------------------------|----------