From 31fa2b6c683b7d7ca2621ccb11d0e2653fe2f5f7 Mon Sep 17 00:00:00 2001
From: Herumb Shandilya <43719685+krypticmouse@users.noreply.github.com>
Date: Tue, 28 Feb 2023 20:42:42 +0530
Subject: [PATCH] [GPTJ] Fix gradient checkpointing bug  (#21794)

* If applied, this commit fixes generate bug in gptj

* Remove extra same code block

* formatting and test fix

* Conflict fix and declaration error fix

---------

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
---
 src/transformers/models/gptj/modeling_gptj.py | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/src/transformers/models/gptj/modeling_gptj.py b/src/transformers/models/gptj/modeling_gptj.py
index 543eb6937fa..5530638c1dd 100755
--- a/src/transformers/models/gptj/modeling_gptj.py
+++ b/src/transformers/models/gptj/modeling_gptj.py
@@ -633,6 +633,13 @@ class GPTJModel(GPTJPreTrainedModel):
 
         output_shape = input_shape + (hidden_states.size(-1),)
 
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
         presents = () if use_cache else None
         all_self_attentions = () if output_attentions else None
         all_hidden_states = () if output_hidden_states else None
@@ -652,11 +659,6 @@ class GPTJModel(GPTJPreTrainedModel):
                 all_hidden_states = all_hidden_states + (hidden_states,)
 
             if self.gradient_checkpointing and self.training:
-                if use_cache:
-                    logger.warning_once(
-                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
-                    )
-                    use_cache = False
 
                 def create_custom_forward(module):
                     def custom_forward(*inputs):