Update Examples to New UX

[Satrat]

Updating existing examples and adding a new example for quantizing a 2:4 sparse model with a custom dataset.

Evaluation

Ran sparseml.evaluate [MODEL] -d wikitext -i lm-evaluation-harness on the output of each example and the baselines, perplexity results below

• base llama2_7b dense ultrachat: 10.10
• llama2_7b w8a8 channelwise weights, dynamic per token activations: 10.17
• llama2_7b w4a16 channelwise weights: 11.09
• baseline llama2_7b cnn_dailymail 2:4 sparsity: 13.35
• llama2_7b 2:4 sparsity w4a16 channelwise weights: 17.31

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB