KasuleTrevor/cdli-qwen3-asr-lg-typical-textpre-0p6b-v2-finetune

CDLI Qwen3-ASR Luganda Typical Speech Fine-tune

This repo contains the selected checkpoint checkpoint-20500 from the LG-QWEN3-ASR-TYPICAL-TEXTPRE-0P6B-V2-STAGE1 run.

Training Setup

• —Base model: KasuleTrevor/cdli-qwen3-asr-lg-textpre-0p6b-v2
• —Datasets: KasuleTrevor/lg_100hrs + dmusingu/yogera-dataset (Luganda split)
• —Training language tag: Luganda
• —Forced inference language: disabled
• —Epochs: 5
• —Batch size: 8
• —Gradient accumulation: 2
• —Learning rate: 2e-05
• —Scheduler: cosine
• —Warmup ratio: 0.03
• —Save steps: 500
• —Selected checkpoint: checkpoint-20500
• —Selection reason: best validation normalized WER (0.218026, CER 0.048474)

Final Test Metrics

• —Corpus WER (normalized): 0.265987
• —Corpus CER (normalized): 0.069419
• —Average utterance WER (normalized): 0.270043
• —Average utterance CER (normalized): 0.070111

Checkpoint Selection Evidence

Source Dataset Breakdown

Artifacts

• —Result folder: results/checkpoint-20500/
• —Includes checkpoint validation summaries, final test predictions, scored outputs, and grouped analyses.