Update README.md

README.md

@@ -7,4 +7,23 @@ language:
 ---
 # Whisper Large RixVox Swedish
 
-This is a Whisper-large finetuned for Swedish using the RixVox dataset.
+This is a [Whisper large](https://huggingface.co/openai/whisper-large-v2) finetuned for Swedish using the [RixVox](https://huggingface.co/datasets/KBLab/rixvox) dataset.
+
+## Evaluation using [Common Voice 11](#):
+
+* WER: XYZ
+* WER (normalized): XYZ
+
+## Training
+
+Training was done using Huggingface and Deepspeed with ZeRO stage 2.
+
+* learning rate: 1e-5
+* optimizer: CPUAdamW (deepspeed)
+* lr scheduler: linear
+* warmup steps: 500
+* per device batch size: 20
+* GPUs: 8 x NVIDIA A100 40GB
+* total batch size: 160
+* steps: 20000
+* entire decoder was frozen