Gemma 2 vs. Pythia

Overview

Gemma 2 succeeds the Gemma family of lightweight open models from Google built using the same processes used for the the larger Gemini models.

Gemma 2 is the successor to the Gemma family of open models, including larer models (9B and 27B parameters) with outsized performance across benchmarks. Using a combination of techniques such as training on twice as much data, knowledge distillation, and architectural improvements such as sliding window attention, logit soft-capping, and model merging, Gemma 2 outperforms models of similar size (such as Llama 3), with the 27B parameter model being competitive with models more than twice its size (such as Llama 70B).

Initial release: 2024-06-27

Reference

https://blog.google/technology/developers/google-gemma-2/

Further Reading

• https://ai.google.dev/gemma/docs
• https://huggingface.co/blog/gemma2

Overview

The most recent (as of May 2023) effort from EleutherAI, Pythia is a set of LLMs trained on The Pile. While it appears to outperform OPT and GPTNeo, its performance against GPT-J is unclear. Versions of Pythia have also been instruct-tuned by the team at Together.

Initial release: 2023-02-13

Reference

https://github.com/EleutherAI/pythia

Further Reading