KasuleTrevor
cdli-qwen3-asr-lg-typical-textpre-1p7b-v2-finetune
Available on FriendliAI
Dedicated EndpointsRun this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
KasuleTrevor
Supported Functionality
Dedicated Endpoints
README
License: apache-2.0
Training Setup
- Base model source: local checkpoint
/workspace/qwen3_asr_lg_textpre_1p7b_v2/merged
- Datasets: KasuleTrevor/lg_100hrs + dmusingu/yogera-dataset (Luganda split)
- Training language tag:
Luganda
- Forced inference language:
Luganda
- Epochs: 10
- Batch size: 4
- Gradient accumulation: 2
- Learning rate: 2e-05
- Scheduler: cosine
- Warmup ratio: 0.03
- Save steps: 500
- Selected checkpoint: checkpoint-25000
- Selection reason: fallback best eval_loss (0.399143)
Final Test Metrics
- Corpus WER (normalized): 0.380703
- Corpus CER (normalized): 0.095308
- Average utterance WER (normalized): 0.384094
- Average utterance CER (normalized): 0.096406
Checkpoint Selection Evidence
Table with columns: checkpoint, step, eval_loss| checkpoint | step | eval_loss |
|---|
| checkpoint-24000 | 24000 | 0.3991716802120209 |
| checkpoint-24500 | 24500 | 0.3993058204650879 |
| checkpoint-25000 | 25000 | 0.3991433680057525 |
| checkpoint-25314 | 25314 | 0.3991433680057525 |
Source Dataset Breakdown
Table with columns: source_dataset, n_samples, mean_wer, mean_cer, median_wer, median_cer| source_dataset | n_samples | mean_wer | mean_cer | median_wer | median_cer |
|---|
| lg100 | 2828 | 0.3841 | 0.0964 | 0.3333 | 0.0652 |
Artifacts
- Result folder:
results/checkpoint-25000/
- Includes checkpoint validation summaries, final test predictions, scored outputs, and grouped analyses.