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

base_model = "mistralai/Mistral-7B-v0.3"
tokenizer = AutoTokenizer.from_pretrained(base_model)
model = AutoModelForCausalLM.from_pretrained(base_model, torch_dtype=torch.float16, device_map="auto")
model = PeftModel.from_pretrained(model, "TatarNLPWorld/mistral-7b-tatar-lora-r8")

prompt = "Минем исемем"
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0]))

📈 Generation Example

Prompt: "Минем исемем"
Generated: "...нең 10 яшьлек аны белән сөйләделәр"

📊 Performance Comparison

Model	Perplexity
Mistral-7B r16	3.98
Mistral-7B r8	4.04
GPT-2 medium	2.91 (full)

👥 Author

Arabov Mullosharaf Kurbonovich

📜 Citation

bibtex

@software{mistral_tatar_lora_r8_2026,
    title = {{Mistral-7B LoRA Adapter for Tatar Language (Rank 8)}},
    author = { Arabov Mullosharaf Kurbonovich},
    year = {2026},
    publisher = {Hugging Face},
    url = {https://huggingface.co/TatarNLPWorld/mistral-7b-tatar-lora-r8}
}