Request to rename the model to "mGPT-1.3B-turkish".
Having conducted extensive tests on the model using 500 different prompts, I have found a significant leaning towards the Turkish language in its output.
Specifically:
Results from the prompts:
About 80% of the time, the model produced content in Turkish. This overwhelming majority suggests a predisposition in its training data towards Turkish.Turkmen language output:
When the model did try to generate content in Turkmen, the output was not accurate and often nonsensical. It's evident that the model struggles with Turkmen, possibly due to inadequate or mislabeled training data.Potential dataset composition:
The model's bias towards Turkish hints that a significant portion of its training data might consist of Turkish content, causing it to perform optimally in Turkish and subpar in Turkmen.
Given these observations, there's a clear case for a name change. By renaming the model to "mGPT-1.3B-turkish", users will have a more accurate expectation of its capabilities and inherent biases.