A 65B "fully open-source" model. This model reaches similar performance to Llama 2 70B and uses less compute (only 1.4 trillion tokens).
Relative Adoption Metric contextualizes downloads against the model's size bucket.
