FunctionGemma PocketAssist 270M
Fine-tuned FunctionGemma 270M for PocketAssist.
Accuracy: 35.0% on 440 examples
| Function |
Accuracy |
| get_password |
0.0% |
| save_password |
80.0% |
| search |
30.0% |
Training
- Base:
unsloth/functiongemma-270m-it
- Method: QLoRA 4-bit (Unsloth) + sequence packing
- LoRA rank: 16, alpha: 32
- LR: 0.0002, epochs: 5
- seq_len: 256, batch: 4×4
- Hardware: Colab T4 (16 GB VRAM)
Usage
from unsloth import FastLanguageModel
model, tokenizer = FastLanguageModel.from_pretrained(
"2796gauravc/functiongemma-pocketassist-270m",
load_in_4bit=True,
)
