mirror of
https://github.com/ml-explore/mlx-examples.git
synced 2025-09-01 04:14:38 +08:00
fix deepseek sharding (#1242)
This commit is contained in:
@@ -386,6 +386,7 @@ class DeepseekV2Model(nn.Module):
|
||||
self.num_layers = layers_per_rank
|
||||
self.layers = self.layers[: self.end_idx]
|
||||
self.layers[: self.start_idx] = [None] * self.start_idx
|
||||
self.num_layers = len(self.layers) - self.start_idx
|
||||
|
||||
def __call__(
|
||||
self,
|
||||
|
Reference in New Issue
Block a user