Optimize model (e.g. deepseek R1) output and remove length restrictions on model output.
closed
d
dimmi
Typically, Merlin's model outputs are more concise compared to the official versions. It's evident that Merlin's developers have placed restrictions on the outputs, which should be removed. Additionally, the outputs need optimization. For instance, the latest Deepseek R1 model produces responses without any visible reasoning process, and its output length is severely limited, resulting in poor performance. We hope to achieve consistency with the official outputs. Given how affordable the Deepseek R1 model is, Merlin developers' current solution makes no sense and significantly impacts user experience.
S
Siddhartha
closed
J
Joey
Siddhartha Did anything change or will this not be adjusted?
S
Siddhartha
Joey We had done improvements to this and adjusted the prompt. Beyond this is not in our control, hence closing the same.
Merlin
under review