Sparse Quantization Example Clarification (#2334)

* clarify example

* cleanup

* update examples

* update output name

Author: Sara Adkins

examples/llama7b_sparse_quantized/README.md

@@ -4,7 +4,8 @@ This example uses SparseML and Compressed-Tensors to create a 2:4 sparse and qua
 The model is calibrated and trained with the ultachat200k dataset.
 At least 75GB of GPU memory is required to run this example.
 
-Follow the steps below, or to run the example as `python examples/llama7b_sparse_quantized/llama7b_sparse_w4a16.py`
+Follow the steps below one by one in a code notebook, or run the full example script 
+as `python examples/llama7b_sparse_quantized/llama7b_sparse_w4a16.py`
 
 ## Step 1: Select a model, dataset, and recipe
 In this step, we select which model to use as a baseline for sparsification, a dataset to
@@ -36,7 +37,8 @@ recipe = "2:4_w4a16_recipe.yaml"
 
 ## Step 2: Run sparsification using `apply`
 The `apply` function applies the given recipe to our model and dataset.
-The hardcoded kwargs may be altered based on each model's needs.
+The hardcoded kwargs may be altered based on each model's needs. This code snippet should 
+be run in the same Python instance as step 1.
 After running, the sparsified model will be saved to `output_llama7b_2:4_w4a16_channel`.
 
 ```python
@@ -67,7 +69,7 @@ apply(
 ### Step 3: Compression
 
 The resulting model will be uncompressed. To save a final compressed copy of the model 
-run the following:
+run the following in the same Python instance as the previous steps.
 
 ```python
 import torch

examples/llama7b_w8a8_quantization.py

@@ -16,12 +16,12 @@
                         num_bits: 8
                         type: "int"
                         symmetric: true
-                        strategy: "channel"
+                        strategy: "tensor"
                     input_activations:
                         num_bits: 8
                         type: "int"
                         symmetric: true
-                        dynamic: True
+                        dynamic: true
                         strategy: "token"
                     targets: ["Linear"]
 """
@@ -37,7 +37,7 @@
 dataset = "ultrachat-200k"
 
 # save location of quantized model out
-output_dir = "./output_llama7b_w8a8_channel_dynamic_compressed"
+output_dir = "./output_llama7b_w8a8_dynamic_compressed"
 
 # set dataset config parameters
 splits = {"calibration": "train_gen[:5%]"}