Unverified Commit e07a3d95 authored by Karim Foda's avatar Karim Foda Committed by GitHub
Browse files

Fix gradient checkpointing bug in git (#21818)


Co-authored-by: default avatarSylvain Gugger <35901082+sgugger@users.noreply.github.com>
parent 50db7414
......@@ -431,6 +431,13 @@ class GitEncoder(nn.Module):
pixel_values_present: Optional[bool] = False,
return_dict: Optional[bool] = True,
) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPast]:
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
all_hidden_states = () if output_hidden_states else None
all_self_attentions = () if output_attentions else None
......@@ -443,11 +450,6 @@ class GitEncoder(nn.Module):
past_key_value = past_key_values[i] if past_key_values is not None else None
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):
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment