feat: Add Gemma 3 12B with QLoRA fine-tuning

- Added PEFT, bitsandbytes, TRL for LoRA training
- 4-bit QLoRA quantization for 48GB GPU fit
- Instruction-tuning format for Gemma chat template
- Auto-detect model type (BERT vs LLM)
- Updated GPU tier to ADA_24/AMPERE_48

components/runpod_trainer/requirements.txt

@@ -1,5 +1,5 @@
 runpod>=1.7.0
-transformers==4.44.0
+transformers>=4.48.0
 datasets>=2.16.0
 accelerate>=0.30.0
 boto3>=1.34.0
@@ -7,3 +7,6 @@ scikit-learn>=1.3.0
 scipy>=1.11.0
 safetensors>=0.4.0
 requests>=2.31.0
+peft>=0.14.0
+bitsandbytes>=0.45.0
+trl>=0.14.0