|
|
--- |
|
|
base_model: unsloth/SmolLM2-1.7B |
|
|
library_name: peft |
|
|
license: apache-2.0 |
|
|
language: |
|
|
- tr |
|
|
metrics: |
|
|
- name: ROUGE-1 |
|
|
type: rouge |
|
|
value: 0.2439 |
|
|
- name: ROUGE-2 |
|
|
type: rouge |
|
|
value: 0.1303 |
|
|
- name: ROUGE-L |
|
|
type: rouge |
|
|
value: 0.2147 |
|
|
- name: BLEU |
|
|
type: bleu |
|
|
value: 0.0406 |
|
|
- name: METEOR |
|
|
type: meteor |
|
|
value: 0.2262 |
|
|
- name: BERTScore Precision |
|
|
type: bertscore |
|
|
value: 0.5286 |
|
|
- name: BERTScore Recall |
|
|
type: bertscore |
|
|
value: 0.5834 |
|
|
- name: BERTScore F1 |
|
|
type: bertscore |
|
|
value: 0.553 |
|
|
--- |
|
|
+---------------------+--------+ |
|
|
| Metrik | Değer | |
|
|
+---------------------+--------+ |
|
|
| ROUGE-1 | 0.2439 | |
|
|
| ROUGE-2 | 0.1303 | |
|
|
| ROUGE-L | 0.2147 | |
|
|
| BLEU | 0.0406 | |
|
|
| METEOR | 0.2262 | |
|
|
| BERTScore Precision | 0.5286 | |
|
|
| BERTScore Recall | 0.5834 | |
|
|
| BERTScore F1 | 0.553 | |
|
|
+---------------------+--------+ |
|
|
|
|
|
✅ Model evaluation is complete and all results are logged to `wandb`. |
|
|
|
|
|
--- |
|
|
|
|
|
# Model Card for SmolLM2-Ziraat-Turkish-v1 |
|
|
|
|
|
<!-- Türkçe versiyonu aşağıda yer almaktadır. --> |
|
|
|
|
|
## 🧠 Model Summary |
|
|
|
|
|
**SmolLM2-Ziraat-Turkish-v1** is a fine-tuned version of the [unsloth/SmolLM2-1.7B](https://huggingface.co/unsloth/SmolLM2-1.7B) model, trained using [Unsloth](https://github.com/unslothai/unsloth) and PEFT (Parameter-Efficient Fine-Tuning). This model has been tailored for Turkish language tasks with a focus on agriculture, finance, and general-purpose conversation. |
|
|
|
|
|
## 🇹🇷 Model Özeti |
|
|
|
|
|
**SmolLM2-Ziraat-Turkish-v1**, [unsloth/SmolLM2-1.7B](https://huggingface.co/unsloth/SmolLM2-1.7B) tabanlı bir model olup, [Unsloth](https://github.com/unslothai/unsloth) ve PEFT (Parameter-Efficient Fine-Tuning) yöntemleriyle Türkçe diline yönelik olarak eğitilmiştir. Tarım, finans ve genel sohbet amaçlı kullanım senaryoları için optimize edilmiştir. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🔍 Model Details / Model Detayları |
|
|
|
|
|
- **Developed by / Geliştiren:** [hosmankarabulut](https://huggingface.co/hosmankarabulut) |
|
|
- **Model type / Model türü:** Causal Language Model (AutoRegressive) |
|
|
- **Language / Dil:** Turkish (Türkçe) |
|
|
- **License / Lisans:** apache-2.0 |
|
|
- **Fine-tuned with / Eğitim Aracı:** [Unsloth](https://github.com/unslothai/unsloth) + PEFT |
|
|
- **Base model / Taban model:** [unsloth/SmolLM2-1.7B](https://huggingface.co/unsloth/SmolLM2-1.7B) |
|
|
|
|
|
--- |
|
|
|
|
|
## 🔗 Sources / Kaynaklar |
|
|
|
|
|
- **Model Repository / Model Deposu:** [https://huggingface.co/hosmankarabulut/SmolLM2-Ziraat-Turkish-v1](https://huggingface.co/hosmankarabulut/SmolLM2-Ziraat-Turkish-v1) |
|
|
- **Base model / Taban model:** [unsloth/SmolLM2-1.7B](https://huggingface.co/unsloth/SmolLM2-1.7B) |
|
|
|
|
|
--- |
|
|
|
|
|
## ✅ Intended Uses / Amaçlanan Kullanım |
|
|
|
|
|
- Turkish chatbots, Q&A systems |
|
|
- Agricultural and financial assistants |
|
|
- General-purpose Turkish LLMs |
|
|
|
|
|
--- |
|
|
|
|
|
## 🚫 Out-of-Scope Use / Uygun Olmayan Kullanım |
|
|
|
|
|
- Medical, legal, or high-risk decision-making |
|
|
- Misinformation or unethical applications |
|
|
|
|
|
--- |
|
|
|
|
|
## ⚠️ Bias, Risks and Limitations / Önyargılar, Riskler ve Sınırlamalar |
|
|
|
|
|
Model may still contain biases inherited from the base model. Performance is best within Turkish language and domain-specific contexts (agriculture, finance). |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧪 Training & Evaluation / Eğitim ve Değerlendirme |
|
|
|
|
|
- **Training Library / Eğitim Kütüphanesi:** [Unsloth](https://github.com/unslothai/unsloth) |
|
|
- **Hardware Used / Kullanılan Donanım:** RTX 3090 |
|
|
- **Precision:** bf16 (mixed precision) |
|
|
- **Dataset:** Özel Türkçe veriseti (tarım odaklı) |
|
|
- **Evaluation Tool:** `wandb` (Weights & Biases) |
|
|
|
|
|
### 📊 Evaluation Results / Değerlendirme Sonuçları |
|
|
|
|
|
| Metric | Value | |
|
|
|----------------------|--------| |
|
|
| ROUGE-1 | 0.2439 | |
|
|
| ROUGE-2 | 0.1303 | |
|
|
| ROUGE-L | 0.2147 | |
|
|
| BLEU | 0.0406 | |
|
|
| METEOR | 0.2262 | |
|
|
| BERTScore Precision | 0.5286 | |
|
|
| BERTScore Recall | 0.5834 | |
|
|
| BERTScore F1 | 0.553 | |
|
|
|
|
|
✅ Tüm metrikler başarıyla hesaplandı ve `wandb` üzerinde kaydedildi. |
|
|
|
|
|
--- |
|
|
|
|
|
## 💡 Quickstart / Hızlı Başlangıç |
|
|
|
|
|
```python |
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
import torch |
|
|
|
|
|
model = AutoModelForCausalLM.from_pretrained("hosmankarabulut/SmolLM2-Ziraat-Turkish-v1", torch_dtype=torch.bfloat16) |
|
|
tokenizer = AutoTokenizer.from_pretrained("hosmankarabulut/SmolLM2-Ziraat-Turkish-v1") |
|
|
|
|
|
inputs = tokenizer("Türkiye'de tarım politikaları hakkında ne düşünüyorsun?", return_tensors="pt") |
|
|
outputs = model.generate(**inputs, max_new_tokens=100) |
|
|
print(tokenizer.decode(outputs[0], skip_special_tokens=True)) |
