---
library_name: peft
pipeline_tag: text-generation
tags:
- lora
- peft
- transformers
- audio-text
- depression-detection
---

# woz checkpoint-130 LoRA adapter

This repository contains the LoRA adapter exported from the best `checkpoint-130`
used in the `Step-Audio2 DAIC-WOZ Depression` experiments.

## Contents

- `adapter_model.safetensors`
- `adapter_config.json`
- `additional_config.json`
- `args.json`
- `dev_metrics.json`

## Intended base model

This adapter is intended to be loaded on top of:

- `Step-Audio2-mini-Think`

The base model is not redistributed in this repository.

## Evaluation snapshot

The included `dev_metrics.json` reports the following result on the project dev
split used during this experiment:

- accuracy: `0.8571`
- weighted_f1: `0.8601`
- macro_f1: `0.8493`
- f1_yes: `0.8148`
- tp: `11`
- tn: `19`
- fp: `4`
- fn: `1`

## Notes

- This repository only packages the inference-relevant adapter artifacts.
- Optimizer state and other training-only checkpoint files are intentionally
  excluded.