# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("oudmaria/darija-translator", dtype="auto")Quick Links
Darija โ English Translator
Fine-tuned Llama 3.2 3B on Moroccan Darija to English translation.
- Base model: unsloth/Llama-3.2-3B-Instruct
- Training data: atlasia/darija_english (doda config, 44,713 pairs)
- Method: QLoRA (4-bit, r=16)
Usage
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
from peft import PeftModel
import torch
tokenizer = AutoTokenizer.from_pretrained("oudmaria/darija-translator")
bnb_config = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_compute_dtype=torch.bfloat16)
base = AutoModelForCausalLM.from_pretrained("unsloth/Llama-3.2-3B-Instruct", quantization_config=bnb_config, device_map="auto")
model = PeftModel.from_pretrained(base, "oudmaria/darija-translator")
messages = [
{"role": "system", "content": "You are a translation assistant. Translate the Moroccan Darija text to English. Reply with ONLY the English translation, nothing else."},
{"role": "user", "content": "Translate: labas 3lik?"},
]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to("cuda")
out = model.generate(input_ids=inputs, max_new_tokens=64)
print(tokenizer.decode(out[0][inputs.shape[1]:], skip_special_tokens=True))
# โ are you okay
Framework versions
- PEFT 0.19.1
- Downloads last month
- 2
Model tree for oudmaria/darija-translator
Base model
meta-llama/Llama-3.2-3B-Instruct Finetuned
unsloth/Llama-3.2-3B-Instruct
# Use a pipeline as a high-level helper # Warning: Pipeline type "translation" is no longer supported in transformers v5. # You must load the model directly (see below) or downgrade to v4.x with: # 'pip install "transformers<5.0.0' from transformers import pipeline pipe = pipeline("translation", model="oudmaria/darija-translator")