Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,86 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
+
license: apache-2.0
|
| 5 |
+
base_model:
|
| 6 |
+
- Esperanto/Medical-Whisper-large-1.5b
|
| 7 |
+
tags:
|
| 8 |
+
- generated_from_trainer
|
| 9 |
+
datasets:
|
| 10 |
+
- medical_data
|
| 11 |
+
- Na0s/Primock_med
|
| 12 |
+
model-index:
|
| 13 |
+
- name: Medical_Whisper_large_1.5b
|
| 14 |
+
results: []
|
| 15 |
+
metrics:
|
| 16 |
+
- cer
|
| 17 |
+
- wer
|
| 18 |
+
pipeline_tag: automatic-speech-recognition
|
| 19 |
+
---
|
| 20 |
+
|
| 21 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
| 22 |
+
should probably proofread and complete it, then remove this comment. -->
|
| 23 |
+
|
| 24 |
+
[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/em-hmh/huggingface/runs/wjd3chz6)
|
| 25 |
+
# Medical_Whisper_large_1.5b
|
| 26 |
+
|
| 27 |
+
This model is a fine-tuned version of [openai/whisper-large-v3](https://huggingface.co/openai/whisper-large-v3) on the primock_data dataset.
|
| 28 |
+
|
| 29 |
+
## Model description
|
| 30 |
+
|
| 31 |
+
Fine tuned version of whisper-large-v3 through transfer learning on Doctor/Patient consultations. This version in the ONNX format in fp32 precision. Stay tuned for instructions on how to run this pipeline in OnnxRuntime!
|
| 32 |
+
|
| 33 |
+
## Intended uses & limitations
|
| 34 |
+
|
| 35 |
+
Medical transcription
|
| 36 |
+
|
| 37 |
+
## Training and evaluation data
|
| 38 |
+
|
| 39 |
+
Na0s/Primock_med
|
| 40 |
+
|
| 41 |
+
## Training procedure
|
| 42 |
+
|
| 43 |
+
Exhaustive transfer learning
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
### Training hyperparameters
|
| 47 |
+
|
| 48 |
+
The following hyperparameters were used during training:
|
| 49 |
+
- learning_rate: 1e-05
|
| 50 |
+
- train_batch_size: 6
|
| 51 |
+
- eval_batch_size: 6
|
| 52 |
+
- seed: 42
|
| 53 |
+
- gradient_accumulation_steps: 4
|
| 54 |
+
- total_train_batch_size: 24
|
| 55 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 56 |
+
- lr_scheduler_type: constant_with_warmup
|
| 57 |
+
- lr_scheduler_warmup_steps: 50
|
| 58 |
+
- training_steps: 500
|
| 59 |
+
- mixed_precision_training: Native AMP
|
| 60 |
+
|
| 61 |
+
### Performance Overview:
|
| 62 |
+
|
| 63 |
+
\| Model Name | WER | CER | Number of Parameters |
|
| 64 |
+
|--------------------|------|------|----------------------|
|
| 65 |
+
| Whisper Tiny | 0.46 | 0.27 | 39M |
|
| 66 |
+
| Whisper Base | 0.42 | 0.26 | 74M |
|
| 67 |
+
| Whisper Small | 0.39 | 0.26 | 244M |
|
| 68 |
+
| Whisper Medium | 0.37 | 0.23 | 769M |
|
| 69 |
+
| Whisper Large v3 | 0.33 | 0.18 | 1.55B |
|
| 70 |
+
| **Whisper Medical**| **0.19** | **0.10** | **1.55B** |
|
| 71 |
+
|
| 72 |
+
**Performance of foundation Whispers vs Medical Whisper on the Validation set.**
|
| 73 |
+
|
| 74 |
+
| Model Name | WER | CER | Number of Parameters |
|
| 75 |
+
|--------------------|------|------|----------------------|
|
| 76 |
+
| **Whisper Medical**| **0.24** | **0.13** | **1.55B** |
|
| 77 |
+
|
| 78 |
+
**Table: Performance of Medical Whisper on the Test set.**
|
| 79 |
+
|
| 80 |
+
|
| 81 |
+
### Framework versions
|
| 82 |
+
|
| 83 |
+
- Transformers 4.42.4
|
| 84 |
+
- Pytorch 2.3.1+cu121
|
| 85 |
+
- Datasets 2.20.0
|
| 86 |
+
- Tokenizers 0.19.1
|