lora-marathi

LoRA-Marathi

Microsoft Phi-2 model fine-tuned on 10k English-Marathi text pairs using LoRA.

Dataset used:

https://huggingface.co/datasets/anujsahani01/English-Marathi

Usage

This repo contains LoRA adapter weights trained on a Marathi → English translation dataset. To use this model, load the base model and then apply this LoRA adapter.

!pip install transformers peft accelerate

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model_id = "your-base-model-here"
lora_model_id = "spnandpure/lora-marathi"

# Load base model
tokenizer = AutoTokenizer.from_pretrained(base_model_id)
model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    device_map="auto",
)

# Load LoRA adapter
model = PeftModel.from_pretrained(model, lora_model_id)

# Example usage
prompt = "Translate from Marathi to English:\nMarathi: मला भूक लागली आहे.\nEnglish:"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))