GSM8K zero-shot (core LLM math capability benchmark) Qwen 3 8b Base: 0.11 Qwen 3 8b Instruct: 0.59 Gradients Instruct 8b (starting from Qwen 3 8b base): 0.68 Yep - you read that right. Training on Grads >> Qwen teams? Full annoucement in Novelty Search next week!
7,25K