[NEW FAMILY OF MODELS] Supra1.5 family just released!

SupraLabs has released the Supra-1.5-50M family, which includes a Base model with a context window expanded from 1,024 to 5,120 tokens, and an Instruct fine-tuned variant, both utilizing RoPE scaling. The training data has been significantly increased to 3 trillion tokens, incorporating diverse sources such as tool calling data and ChatML conversations. This release is crucial for practitioners as it offers enhanced context handling and improved performance on various tasks, particularly in raw inference scenarios, making it a valuable asset for fine-tuning and reinforcement learning applications.