From c0c063edbf03713d853f3ca2ed0a26643fbbc76f Mon Sep 17 00:00:00 2001 From: alugupta Date: Mon, 25 Jun 2018 15:18:41 -0400 Subject: [PATCH] Update README.md --- README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 04537b1..665ae02 100644 --- a/README.md +++ b/README.md @@ -30,9 +30,9 @@ The MNIST and CiFAR datasets are made available through the Keras deep learning [TIDIGITs](https://catalog.ldc.upenn.edu/ldc93s10) and [ImageNet](http://www.image-net.org/) must be downloaded and pre-processed separately, however pre-trained models for ImageNet (e.g., VGG16, ResNet50) are available through Keras as well. ### Quantization -To quantize models, run `ares/experiments/quantize/run.sh`. +To quantize models, run [`ares/experiments/quantize/run.sh`](./experiments/quantize/run.sh). -The quantization transform [`ares/dl_models/transform/quantize.py`](./dl-models/transform/quantize.py) models fixed-point datatypes for weights. Note that activations and arithmetic operations still have full-precision. +The [quantization transform](./dl-models/transform/quantize.py) emulates fixed-point datatypes for weights. Note that activations and arithmetic operations still have full-precision. ### Evaluation After training and quantizing models, `ares/experiments/eval/eval.sh` can be used to evaluate the models on the validation and test sets.