UPdate readme

README.md

@@ -1,6 +1,6 @@
 This model is a downstream optimization of [```vuiseng9/bert-base-squadv1-block-pruning-hybrid-filled-lt```](https://huggingface.co/vuiseng9/bert-base-squadv1-block-pruning-hybrid-filled-lt) using [OpenVINO/NNCF](https://github.com/openvinotoolkit/nncf). Applied optimization includes:
-1. magnitude sparsification during initialization
-2. NNCF Quantize-Aware Training
+1. magnitude sparsification at 50% upon initialization. Parameters are ranked globally via thier absolute norm. Only linear layers of self-attention and ffnn are targeted.
+2. NNCF Quantize-Aware Training - Symmetric 8-bit for both weight and activation on all learnable layers.
 3. Custom distillation with large model ```bert-large-uncased-whole-word-masking-finetuned-squad```  
 
 ```
@@ -84,7 +84,6 @@ python run_qa.py \
 ```
 
 # Eval
-
 This repo must be cloned locally.
 ```bash
 git clone https://huggingface.co/vuiseng9/bert-base-squadv1-block-pruning-hybrid-filled-lt-nncf-50.0sparse-qat-lt