Issue with generating Czech character 'ý'
#6
by
mstrunc
- opened
Hello, I've found that the deepseek-llm-7b-chat model has an issue generating the Czech character 'ý'. When prompted in Czech, the model's output contains a replacement character (�, bytes \xef\xbf\xbd) instead of 'ý'. Other Czech characters seem to work fine. I have confirmed that the issue is specific to this model, as other models like Llama 3 8B work correctly in the same environment. This suggests a problem within the model's vocabulary or training for this specific character.