From 6443f2e8e297ec5d7a6938098eb015d3a789b76a Mon Sep 17 00:00:00 2001
From: Dan Saattrup Nielsen <dan.nielsen@alexandra.dk>
Date: Tue, 29 Oct 2024 09:34:57 +0100
Subject: [PATCH] chore: Add whisper dropout to large turbo, up lr to 3e-5

---
 config/model/whisper-large-turbo.yaml | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/config/model/whisper-large-turbo.yaml b/config/model/whisper-large-turbo.yaml
index a4ad40d6..1ccc9116 100644
--- a/config/model/whisper-large-turbo.yaml
+++ b/config/model/whisper-large-turbo.yaml
@@ -9,8 +9,8 @@ clean_text: true
 
 # Model hyperparameters
 sampling_rate: 16_000
-dropout: 0.0
-activation_dropout: 0.1
+dropout: 0.1
+activation_dropout: 0.0
 attention_dropout: 0.0
 mask_time_prob: 0.5
 mask_time_length: 10
@@ -20,4 +20,4 @@ layerdrop: 0.1  # NOTE: This will automatically be set to 0 in a multi-gpu setti
 max_length: 225
 
 # Model-specific optimisation hyperparameters
-learning_rate: 1e-5
+learning_rate: 3e-5