align torch and keras gptq tutorials #1123
Conversation
| "This is achieved through an optimization process applied post-quantization, specifically adjusting the rounding of quantized weights.\n", | ||
| "GPTQ is especially effective after mixed precision quantization. \n", |
There was a problem hiding this comment.
I think this would be more accurate: GPTQ is especially effective in case of low bit width quantization and mixed precision quantization.
| "collapsed": false | ||
| }, | ||
| "id": "2c13aff20d208c51" | ||
| }, | ||
| { |
There was a problem hiding this comment.
Unify the sequence of code sections into single section..
| " os.makedirs(target_dir / lbl)\n", | ||
| " \n", | ||
| " for img_file, lbl in zip(sorted(os.listdir(imgs_dir)), labels):\n", | ||
| " shutil.move(imgs_dir / img_file, target_dir / lbl)\n" | ||
| ], |
There was a problem hiding this comment.
I've already added such imagenet extractor to: tutorials/notebooks/imx500_notebooks/keras/example_keras_mobilenetv2_for_imx500.ipynb
we need to align.
| "\n", | ||
| "Please ensure that the dataset path has been set correctly." | ||
| "In this tutorial we use the same representative dataset for both statistics collection and GPTQ. A complete pass through the representative dataset generator constitutes an epoch (batch_size x n_iter samples). In this example we use the same dataloader iterator for all epochs, i.e. different images are used in different epochs." | ||
| ] |
There was a problem hiding this comment.
It's a good explanation, but I'd consider to make it shorter or divide it to bullets:
subtitle: representative dataset
Create a generator for the representative dataset ....
A word on GPTQ representative dataset:
- GPTQ is a gradient-base...
- In this tutorial we use the same representative dataset..
| ] | ||
| }, | ||
| { | ||
| "cell_type": "code", | ||
| "execution_count": null, | ||
| "outputs": [], | ||
| "source": [ | ||
| "from model_compression_toolkit.gptq.common.gptq_constants import REG_DEFAULT\n", | ||
| "# Create a GPTQ quantization configuration and set the number of training iterations. \n", |
There was a problem hiding this comment.
I think it will be good to move "import model_compression_toolkit" to this section, to emphasize that we start using the MCT here.
| " !wget https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_val.tar\n", | ||
| " !mv ILSVRC2012_img_val.tar imagenet/" | ||
| " !wget -P imagenet https://image-net.org/data/ILSVRC/2012/ILSVRC2012_devkit_t12.tar.gz\n", | ||
| " !wget -P imagenet https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_val.tar" | ||
| ], |
There was a problem hiding this comment.
very nice! same textual comments for the pythorch..
|
@irenaby |
Pull Request Description:
Checklist before requesting a review: