mirror of
https://github.com/KoboldAI/KoboldAI-Client.git
synced 2025-06-05 21:59:24 +02:00
Breakmodel's CUDA stream should be on primary device
This commit is contained in:
@ -387,7 +387,7 @@ def new_forward(
|
||||
all_hidden_states = () if output_hidden_states else None
|
||||
|
||||
if breakmodel and ram_blocks:
|
||||
copystream = torch.cuda.Stream(device=0,priority = -1)
|
||||
copystream = torch.cuda.Stream(device=primary_device, priority=-1)
|
||||
|
||||
for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
|
||||
|
||||
|
Reference in New Issue
Block a user