Qwen3-ASR-1.7B fine-tuned on ViMD (LoRA)

LoRA adapter (r=16, alpha=32, dropout 0.05) on all attention and MLP projections of the LLM; 17.4M trainable parameters (0.85%). lr 1e-4, AdamW. Training took 62 min on one 24 GB GPU (peak 13.1 GB).

Usage

import torch
from peft import PeftModel
from qwen_asr import Qwen3ASRModel   # pip install qwen-asr

w = Qwen3ASRModel.from_pretrained("Qwen/Qwen3-ASR-1.7B", dtype=torch.bfloat16, device_map="cuda:0")
w.model = PeftModel.from_pretrained(w.model, "duclvQ/qwen3-asr-1.7b-vimd-lora").merge_and_unload().eval()
print(w.transcribe(audio="audio.wav", language="Vietnamese")[0].text)

Results

Held-out test split of ViMD (Vietnamese dialect speech, North + Central regions; 1,368 clips, 7.2 h, 939 speakers, none seen in training). WER/CER are computed after lowercasing and stripping punctuation, with language="Vietnamese" forced.

Training

• Base: Qwen/Qwen3-ASR-1.7B. Data: ViMD North + Central train split, 10,346 clips (~55 h), clips <= 30 s. Transcripts are verbatim: numbers spelled out, fillers and repetitions kept.
• Target format: language Vietnamese<asr_text>{transcript}. 2 epochs (1,294 steps), effective batch 16, cosine schedule, bf16, gradient checkpointing, seed 42.

Caveats

• One seed and one hyperparameter setting per method; differences below ~0.5 WER points are not reliable.
• Part of the improvement over the base model is format learning: the base model writes digits (e.g. "2022") where the references spell numbers out, and handles fillers differently. Excluding digit clips and fillers, the neutralised gain for LoRA is 8.51 -> 6.69 WER (about 64% of the raw gain).
• Evaluated on ViMD only (read/spontaneous dialect speech); no claims about other domains. Check the ViMD dataset license before any commercial use of this model.
• Base model license: Apache-2.0 (Qwen3-ASR).