---
library_name: transformers
tags:
- trl
- cpt
datasets:
- canbingol/vngrs-web-corpus-200k
language:
- tr
- en
base_model:
- google/gemma-3-1b-pt
---

# Model Card: Gemma3-1B Turkish CPT (15K Subset, 3 Epoch)

## Overview

This model is a Turkish Continued Pretraining (CPT) variant of `google/gemma-3-1b-pt`.

The base model was further trained for **3 epochs** on the first **15,000 samples** of a Turkish web corpus to improve Turkish language modeling capability and domain familiarity.

This release is intended for **research and experimental use**.

---

## Base Model

- `google/gemma-3-1b-pt`

---

## Training Setup

- Dataset: `canbingol/vngrs-web-corpus-200k`
- Subset Used: First 15,000 samples
- Training Objective: Continued Pretraining (Causal LM / Next-Token Prediction)
- Epochs: 3
- Data Regime: Plain text (no instruction formatting)
- Token Count (approximate): ~19.5M tokens
---


## Citation

If you use this model, please cite the base model:

- Base: `google/gemma-3-1b-pt`

---

## Usage Example

```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "canbingol/gemma3_1B_base-tr-cpt-3epoch_15k_data"

device = "cuda" if torch.cuda.is_available() else "cpu"

model = AutoModelForCausalLM.from_pretrained(model_id)
tokenizer = AutoTokenizer.from_pretrained(model_id)

model = model.to(device)

prompt = "Benim adım"
inputs = tokenizer(prompt, return_tensors="pt").to(device)

outputs = model.generate(
    **inputs,
    max_new_tokens=50,
    do_sample=True,
    temperature=0.8,
    top_p=0.9
)

generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(generated_text)