Gemma 4 26B-A4B Pruned — Türkçe LoRA Adapter

Bu adapter esokullu/gemma4-tr-26b-a4b-pruned (pruned bf16) base'i üzerinde eğitildi. Merged versiyon için: esokullu/gemma4-tr-26b-a4b-pruned.

Eğitim

LoRA r=32, α=64, dropout=0
target_modules: "all-linear"
2 epoch, 25k sample (TR/code/math mix)
A100 80GB, ~2 saat

Kullanım

```python from peft import PeftModel from transformers import AutoModelForCausalLM import torch

base = AutoModelForCausalLM.from_pretrained( "esokullu/gemma4-tr-26b-a4b-pruned", torch_dtype=torch.bfloat16, device_map="auto", ) model = PeftModel.from_pretrained(base, "esokullu/gemma4-turkish-26b-a4b-pruned-lora") ```

Downloads last month: 8

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support