OxxoCodes/Pula-8B-v0.1

Text Generation • Updated 17 days ago • 4

Note The current (but not final) version of Pula-8B. Currently recommended for most use cases. Excels in translation and reasoning tasks.

OxxoCodes/Pula-XLMR-large-v0.1

Fill-Mask • Updated 17 days ago • 24

Note The current (but not final) version of Pula-XLMR-Large. Currently recommended for most use cases.

Note Instruction-tuning dataset consisting of existing instruction datasets, datasets altered to a chat form, translation data, GPT-4o and Gemini-1.5-Pro translated subsets of popular instruction datasets, and purely synthetic instruction data.

Note Contains raw webtext and other documents written in Setswana or code-switched Setswana+English

Note The entirety of the MMLU test split, translated into Setswana with GPT-4o

Note The entirety of the GSM8K test split, translated into Setswana with Gemini 1.5 Pro.