Instructions to use erhanalsr/langusta-kpss-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use erhanalsr/langusta-kpss-lora with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("unsloth/qwen2.5-0.5b-instruct-unsloth-bnb-4bit") model = PeftModel.from_pretrained(base_model, "erhanalsr/langusta-kpss-lora") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Unsloth Desktop
LangUsta KPSS LoRA Adapter
PEFT adapter trained on erhanalsr/langusta-kpss-reasoning with Unsloth 4-bit QLoRA. This repository contains adapter weights, not a merged full model.
Benchmark Results
TR-MMLU
The adapter and base model were evaluated on 6,200 Turkish MMLU questions.
| Model | Correct | Questions | Accuracy |
|---|---|---|---|
| Qwen2.5-0.5B-Instruct | 1,767 | 6,200 | 28.50% |
| LangUsta-KPSS-LoRA | 1,258 | 6,200 | 20.29% |
LangUsta Custom Benchmark
The custom benchmark contains 100 held-out Turkish five-choice questions.
| Model | Correct | Accuracy |
|---|---|---|
| Qwen2.5-1.5B-Instruct | 33/100 | 33% |
| LangUsta-MCQ-Letter-LoRA | 29/100 | 29% |
| SmolLM2-1.7B-Instruct | 23/100 | 23% |
| Qwen2.5-0.5B-Instruct | 19/100 | 19% |
| Gemma-3-1B-Instruct-4bit | 19/100 | 19% |
- Downloads last month
- 55