Fix model forward grad (#628)
This commit is contained in:
@@ -601,6 +601,7 @@ class Grok1ModelForCausalLM(nn.Module):
|
||||
# Monkey patch _prepare_weights to load pre-sharded weights
|
||||
setattr(DefaultModelLoader, "_prepare_weights", _prepare_presharded_weights)
|
||||
|
||||
@torch.no_grad()
|
||||
def forward(
|
||||
self,
|
||||
input_ids: torch.Tensor,
|
||||
|
||||
Reference in New Issue
Block a user