Suggestions
Sharad Varshney
Generative AI | Fine tuning LLMs | Co-Founder
Sharad Varshney is a seasoned professional with over 18 years of expertise in Machine Learning, AI, and DNN architectures, with a strong background in platform and product development.
Having studied Machine Learning at Stanford University and obtained an MS in Computer Science from California State University-Sacramento, Sharad has a solid educational foundation.
Throughout his career, he has held key roles in renowned organizations such as Epsilon, where he served as the Senior Vice President of AI & ML Engineering, Vice President, Head of AI/ML Engineering, and Head of AI/ML at Cosmos, a Startup Venture within Epsilon.
Sharad's professional journey also includes founding positions at companies like Peritus.ai and Spendwise, as well as significant contributions at Intel Corporation, Global Technology, Getty Images, and GMAC/Residential Capital Corporation.
His specialties encompass a wide array of areas including Machine Learning, Deep Learning using PyTorch and Tensorflow, Neural nets, Large Language Models, Speech2Text, and Object Detection.
Sharad Varshney's depth of experience and leadership in AI and ML make him a distinguished figure in the technology and engineering landscape.
Highlights
Llama 3.1 models, particularly the 405B parameter variant, have been released today as open-source models designed to compete directly with best closed source LLMs of the worlds such as gpt-4o and similar high-end AI models. Llama 3.1 is an auto-regressive language model with an optimized transformer architecture, using SFT and RLAIF for alignment.
For me this is most important open source announcement as this model has closed the gap between the best closed source model and open source models and with change in license terms, we can also use the generated outputs to train or fine tune our own models.
Thanks Meta for their pioneering contributions to the open-source AI community and pushing boundaries like never before.
Some of the key features of this new llama3.1-405B model:
-
Extended Context Length: 128K context length
-
Multilingual Capabilities: Enhanced support across eight languages
-
New Workflows Enabled: From synthetic data generation to model distillation, the possibilities are limitless.
-
Enhanced Safety Tools: Introducing Llama Guard 3, Prompt Guard, and CyberSecEval 3 to ensure responsible development.
Looking to make some connections in NCR and Bengaluru region in the startup community where founder are solving real world problems and possibly looking to raise an early stage investment. If you are at the intersection of solving real world problems using AI - reach out to me with details of what problem you are solving and where you are in your journey. I can help with either some angel investment or with my time. #gurgaonstartups #StartupsInIndia #AIIndianStartups #IndianStartups
