OmoTalk-3B-V1
Built by Maxxcore Research Lab. Fine-tuned on 300,000 sentences across 6 African languages.
Language Quality (V1)
| Language | Status | Notes |
|---|---|---|
| Swahili | ✅ Good | Consistent output |
| Somali | ✅ Good | Solid performance |
| Nigerian Pidgin | ⚠️ Partial | Some English noise |
| Hausa | ⚠️ Partial | Mixed results |
| Yorùbá | 🔬 Experimental | Data cleaned in V2 |
| Igbo | 🔬 Experimental | Data cleaned in V2 |
About
Base model: meta-llama/Llama-3.2-3B Method: Continued pretraining with LoRA + Unsloth Training: 300k sentences, 53 minutes on Kaggle T4 Research lab: maxxcore.ai
This is V1. V2 will include cleaner datasets and instruction fine-tuning for all 6 languages.
- Downloads last month
- 28
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for lintonsui/OmoTalk-3B-V1
Base model
meta-llama/Llama-3.2-3B