KasuleTrevor/cdli-qwen3-asr-lg-typical-textpre-0p6b-finetune-t2

CDLI Qwen3-ASR Luganda Typical Speech Fine-tune

This repo contains the selected checkpoint checkpoint-14300 from the LG-QWEN3-ASR-TYPICAL-TEXTPRE-0P6B-T2 run.

Training Setup

• —Base model: KasuleTrevor/cdli-qwen3-asr-lg-typical-textpre-0p6b-finetune
• —Datasets: KasuleTrevor/lg_100hrs + dmusingu/yogera-dataset (Luganda split)
• —Training language tag: Luganda
• —Forced inference language: disabled
• —Epochs: 10
• —Batch size: 4
• —Gradient accumulation: 2
• —Learning rate: 2e-05
• —Scheduler: cosine
• —Warmup ratio: 0.03
• —Save steps: 500
• —Selected checkpoint: checkpoint-14300
• —Selection reason: fallback best eval_loss (0.392313)

Final Test Metrics

• —Corpus WER (normalized): 0.337463
• —Corpus CER (normalized): 0.088204
• —Average utterance WER (normalized): 0.340240
• —Average utterance CER (normalized): 0.089057

Checkpoint Selection Evidence

Source Dataset Breakdown

Artifacts

• —Result folder: results/checkpoint-14300/
• —Includes checkpoint validation summaries, final test predictions, scored outputs, and grouped analyses.