Fix dataset composition percentages and token counts

Fix dataset mix to 50-30-20 and update code snippet with better generation params

Fix citation to reference blog post

Update README.md

Update README.md

Update README.md

Add comprehensive model card with benchmark results