---
tags: [text-to-sql, tpc-ds, qlora, phi-1_5, retail, jetson, 4bit]

pipeline_tag: text-generation
library_name: transformers
---
# Phi-1.5 TPC-DS SQL QLoRA — v2 (Jetson)
- Base model  : microsoft/phi-1_5
- Method      : QLoRA (4-bit NF4 + fp16 compute, bitsandbytes 0.48.0.dev0)
- Device      : NVIDIA Jetson Orin Nano 8GB / JetPack 6.2
- Samples     : 1000 | Epochs: 5 | LR: 0.0002
- BLEU        : 0.31% | Exact Match: 0.0%
- Trained     : 2026-06-09T21:06:58.536747

## Load this adapter
```python
from transformers import AutoModelForCausalLM, BitsAndBytesConfig
from peft import PeftModel
import torch

bnb = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4",
                          bnb_4bit_compute_dtype=torch.float16)
base = AutoModelForCausalLM.from_pretrained(
    "microsoft/phi-1_5", quantization_config=bnb,
    device_map={"": 0}, trust_remote_code=True)
model = PeftModel.from_pretrained(base, "InfometryINC/phi15-retail-sql-lora-jetson-v2")
```