Vietnamese dual-field translator for the BIRD Text-to-SQL benchmark. Fine-tuned from Qwen/Qwen2.5-7B-Instruct with QLoRA (4-bit NF4) to translate both the question and evidence fields from English into Vietnamese in a single pass.

Used to produce hoadm/vibird.

Model details


Base model	`Qwen/Qwen2.5-7B-Instruct`
Method	QLoRA (4-bit NF4, bfloat16 compute)
LoRA rank	16
LoRA alpha	32
Target modules	q/k/v/o/gate/up/down proj
Training data	3,772 hybrid items (human + GPT, BIRD train)
Epochs	3
Optimizer	paged_adamw_8bit

Usage

python

from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
from peft import PeftModel
import torch

base  = "Qwen/Qwen2.5-7B-Instruct"
adapter = "hoadm/qwen25-bird-translator-vi"

bnb = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_quant_type="nf4",
)
tokenizer = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, quantization_config=bnb, device_map="auto")
model = PeftModel.from_pretrained(model, adapter)
model.eval()

def translate(question: str, evidence: str) -> dict:
    system = (
        "You are an expert translator. Translate the English Text-to-SQL question "
        "and evidence into Vietnamese. Return JSON: "
        '{"question_vi": "...", "evidence_vi": "..."}'
    )
    user = f'Question: {question}\nEvidence: {evidence if evidence.strip() else "(no evidence)"}'
    messages = [{"role": "system", "content": system},
                {"role": "user",   "content": user}]
    text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    inputs = tokenizer(text, return_tensors="pt").to(model.device)
    with torch.no_grad():
        out = model.generate(**inputs, max_new_tokens=256, do_sample=False)
    return tokenizer.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True)

print(translate(
    "What is the highest eligible free rate for K-12 students in Alameda County?",
    "Eligible free rate for K-12 = `Free Meal Count (K-12)` / `Enrollment (K-12)`"
))

Dataset: hoadm/vibird
Sister model (Spider): hoadm/qwen25-spider-translator-vi

Citation

bibtex

@article{li2024bird,
  title   = {Can LLM Already Serve as a Database Interface? A Big Bench for Large-Scale Database Grounded Text-to-SQLs},
  author  = {Li, Jinyang and Hui, Binyuan and Qu, Ge and Yang, Jiaxi and others},
  journal = {Advances in Neural Information Processing Systems},
  volume  = {36},
  year    = {2024},
  url     = {https://arxiv.org/abs/2305.03111}
}

hoadm/qwen25-bird-translator-vi

qwen25-bird-translator-vi

Model details

Usage

Related

Citation

hoadm/
qwen25-bird-translator-vi