---
library_name: peft
base_model: Qwen/Qwen3.5-9B
pipeline_tag: text-generation
language:
- vi
tags:
- lora
- qlora
- text-to-sql
---

# DataMate Qwen3.5 LoRA — checkpoint 275

LoRA weights learned on Vietnamese business-data tool-calling conversations.
This repository contains the trained adapter and tokenizer/processor assets;
the base model is Qwen/Qwen3.5-9B and must be loaded alongside the adapter.

Runtime code supports BM25 retrieval on question text (top 3), three sampled
ReAct rollouts, real SQL execution, majority voting, ETA and agreement counters.
Training examples, source business training data, optimizer state and credentials
are not included. This is a saved training checkpoint, not a claim of best accuracy.
No end-to-end accuracy metric has been established for this release.