From af556a09f67ad279f14e9bcb93780d45a2b68a28 Mon Sep 17 00:00:00 2001
From: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>
Date: Tue, 18 Oct 2022 18:25:49 +0200
Subject: [PATCH] add `accelerate` support for `Whisper` (#19697)

---
 src/transformers/models/whisper/modeling_whisper.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/src/transformers/models/whisper/modeling_whisper.py b/src/transformers/models/whisper/modeling_whisper.py
index 41fab31493..e9d774b562 100644
--- a/src/transformers/models/whisper/modeling_whisper.py
+++ b/src/transformers/models/whisper/modeling_whisper.py
@@ -446,6 +446,7 @@ class WhisperPreTrainedModel(PreTrainedModel):
     base_model_prefix = "model"
     main_input_name = "input_features"
     supports_gradient_checkpointing = True
+    _no_split_modules = ["WhisperEncoderLayer"]
 
     def _init_weights(self, module):
         std = self.config.init_std