[!NOTE] This is a copy of chengyili2005/whisper-small-mandarin-lora at commit
892d846, made on 2026-10-30. vLLM CI uses this checkpoint and the original is on a personal account, so we keep a copy here to reduce supply-chain risk.
Whisper Small Mandarin - Chengyi Li
This model is a fine-tuned version of openai/whisper-small on the Common Voice 24.0 - Mandarin dataset. It achieves the following results on the evaluation set:
- Loss: 0.4541
- Cer: 12.8966
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 40000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer |
|---|---|---|---|---|
| 0.4032 | 0.5411 | 1000 | 0.3531 | 17.9573 |
| 0.2965 | 1.0823 | 2000 | 0.3232 | 14.1834 |
| 0.2964 | 1.6234 | 3000 | 0.3165 | 14.0472 |
| 0.1811 | 2.1645 | 4000 | 0.3155 | 14.5741 |
| 0.1665 | 2.7056 | 5000 | 0.3166 | 13.7639 |
| 0.097 | 3.2468 | 6000 | 0.3293 | 13.4425 |
| 0.1235 | 3.7879 | 7000 | 0.3343 | 13.7206 |
| 0.0801 | 4.3290 | 8000 | 0.3432 | 13.6987 |
| 0.0567 | 4.8701 | 9000 | 0.3471 | 13.6069 |
| 0.0258 | 5.4113 | 10000 | 0.3568 | 13.4482 |
| 0.0374 | 5.9524 | 11000 | 0.3621 | 13.8458 |
| 0.0222 | 6.4935 | 12000 | 0.3677 | 13.1159 |
| 0.0078 | 7.0346 | 13000 | 0.3809 | 13.1915 |
| 0.0104 | 7.5758 | 14000 | 0.3765 | 13.2965 |
| 0.0083 | 8.1169 | 15000 | 0.3826 | 13.7172 |
| 0.0071 | 8.6580 | 16000 | 0.3888 | 13.1649 |
| 0.0036 | 9.1991 | 17000 | 0.3906 | 13.0962 |
| 0.0047 | 9.7403 | 18000 | 0.3964 | 13.2278 |
| 0.0013 | 10.2814 | 19000 | 0.4014 | 13.0939 |
| 0.0026 | 10.8225 | 20000 | 0.4034 | 13.0489 |
| 0.0039 | 11.3636 | 21000 | 0.4064 | 13.1903 |
| 0.0021 | 11.9048 | 22000 | 0.4081 | 13.1568 |
| 0.0009 | 12.4459 | 23000 | 0.4113 | 13.0016 |
| 0.0014 | 12.9870 | 24000 | 0.4138 | 13.0714 |
| 0.0015 | 13.5281 | 25000 | 0.4203 | 13.1482 |
| 0.0009 | 14.0693 | 26000 | 0.4177 | 13.3778 |
| 0.001 | 14.6104 | 27000 | 0.4233 | 12.9970 |
| 0.0004 | 15.1515 | 28000 | 0.4278 | 12.8977 |
| 0.0005 | 15.6926 | 29000 | 0.4298 | 13.1297 |
| 0.0002 | 16.2338 | 30000 | 0.4328 | 12.9272 |
| 0.0002 | 16.7749 | 31000 | 0.4322 | 12.9641 |
| 0.0001 | 17.3160 | 32000 | 0.4368 | 12.9924 |
| 0.0015 | 17.8571 | 33000 | 0.4414 | 13.1222 |
| 0.0001 | 18.3983 | 34000 | 0.4429 | 12.9358 |
| 0.0001 | 18.9394 | 35000 | 0.4459 | 12.9451 |
| 0.0001 | 19.4805 | 36000 | 0.4510 | 12.9658 |
| 0.0001 | 20.0216 | 37000 | 0.4505 | 12.8741 |
| 0.0001 | 20.5628 | 38000 | 0.4511 | 12.8960 |
| 0.0001 | 21.1039 | 39000 | 0.4533 | 12.8735 |
| 0.0003 | 21.6450 | 40000 | 0.4541 | 12.8966 |
Framework versions
- PEFT 0.18.1
- Transformers 4.52.0
- Pytorch 2.9.1+cu128
- Datasets 4.5.0
- Tokenizers 0.21.4