Llama-3.2-1B-Instruct-Hindi

Uploaded model

• Developed by: Ryder99
• License: apache-2.0
• Finetuned from model : unsloth/Llama-3.2-1B-Instruct-unsloth-bnb-4bit

---

This model was created as part of a project for my NLP course at University. I'm limited to what free Colab can handle, but I'm satisfied with the performance of this model considering the small size. With some preliminary testing, it appears to be slightly faster than the base Meta Llama 3.2 1B, while producing significantly better Hindi output and comparable English output. The model appears to prefer outputting Hindi regardless of prompt language if there is any Hindi in the context.

I intend to do some better testing in the future, but until then this model is provided as is for you to try. It might be convenient to use Google Translate or some similar service to write the prompt in Devnagari and translate the output back to English for quick testing if, like me, Hindi is not your first language and you struggle to read it. The model seems viable as an on-device model for Hindi-speakers, with usable inference speeds even on a phone (tested using Ollama on Termux).

A slightly bigger model can be found at Llama-3.2-3B-Instruct-Hindi.

For queries, or to help me train larger models ( ...I couldn't get lab access :( ... ) you can mail me here.

---

This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.

<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>