Update README.md

README.md

@@ -10,11 +10,23 @@ datasets:
 metrics:
 - WER
 ---
+## Model
+* Name: Whisper Large-v2 Swahili
+* Description: Fine-tuned Whisper weights for speech-to-text task.
+* Dataset:
+  - Train and validation splits for Swahili subsets of [Common Voice 11.0](https://huggingface.co/datasets/mozilla-foundation/common_voice_11_0).
+  - Train, validation and test splits for Swahili subsets of [Google Fleurs](https://huggingface.co/datasets/google/fleurs/). 
+* Performance: **19.887087 WER**
 
+## Weights
+* Date of release: 12.09.2022
+* Size:
+* License: MIT
+
+## Usage
+To use these weights in HuggingFace's `transformers` library, you can do the following:
+```python
+from transformers import WhisperForConditionalGeneration
 Trained on: 
 * _"train+validation"_ Swahili subsets of [Common Voice 11.0](https://huggingface.co/datasets/mozilla-foundation/common_voice_11_0)
 * _"train+validation+test"_ Swahili subsets of [Google Fleurs](https://huggingface.co/datasets/google/fleurs/).
-
-|GPU|Train Epochs|Training Loss|Validation Loss|WER|Train Time| Train Batch Size | Validation Batch Size
-|---|----------------|---------------|-------------|----------|---|---|---|
-| 1 x A100 | 1000 | 0.082000 |0.352631| 19.887087 | 11.2 hrs | 32 | 8 |