import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

BASE_MODEL = "CohereLabs/c4ai-command-r7b-arabic-02-2025"
ADAPTER_ID = "moncef-blm/Algerian-Dialect-Conversational-System-for-Smoking-Reduction"

tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
base = AutoModelForCausalLM.from_pretrained(
BASE_MODEL,
device_map="auto",
torch_dtype=torch.float16,
)

model = PeftModel.from_pretrained(base, ADAPTER_ID)
model.eval()

Chat template: use the base model tokenizer's apply_chat_template(...) (same as base).

Training details

Method: PEFT / LoRA (adapter training)
Domain: Algerian Arabic / smoking cessation assistant
Data: The fine-tuning data is primarily in Algerian Arabic (Darja) + MSA, collected from multiple sources (mixed domains and writing styles). Some portions were originally in other languages and were translated into Algerian dialect to expand coverage and increase instruction diversity; translations may introduce artifacts and occasional unnatural phrasing, so outputs should be evaluated accordingly

Limitations / Safety

This is a community fine-tune and may produce incorrect or unsafe outputs. Use with appropriate evaluation and safeguards.

License

This adapter is shared under cc-by-nc-4.0 and inherits constraints from the base model license/terms. Please also follow the base model’s usage requirements.