TasaractDavut Emre Taşar
Go back
Model

Qwen 0.5B · GRPO maths

Wetin e dey do

Small model fit learn how to solve maths word problem and still arrange di answer well? Dis Qwen model get half a billion parameters and e train for one pass through GSM8K. E dey write di solution, then put di final answer for separate section. During GRPO training, different rewards check di number and di format separately. Even solution wey look neat fit still get calculation error.

Original documents

Original source · huggingface.co Qwen2.5-0.5B-Instruct
How to cite am

Qwen 0.5B · GRPO maths · https://huggingface.co/emre/Qwen-0.5B-GRPO