mirror of
https://github.com/KoboldAI/KoboldAI-Client.git
synced 2025-06-05 21:59:24 +02:00
Fix multigpu loading without lazy-loader
This commit is contained in:
@@ -243,6 +243,11 @@ class GenericHFTorchInferenceModel(HFTorchInferenceModel):
|
|||||||
)
|
)
|
||||||
shutil.rmtree("cache/")
|
shutil.rmtree("cache/")
|
||||||
|
|
||||||
|
if not self.lazy_load:
|
||||||
|
utils.layers_module_names = utils.get_layers_module_names(self.model)
|
||||||
|
utils.module_names = list(self.model.state_dict().keys())
|
||||||
|
utils.named_buffers = list(self.model.named_buffers(recurse=True))
|
||||||
|
|
||||||
self.patch_embedding()
|
self.patch_embedding()
|
||||||
|
|
||||||
if utils.koboldai_vars.hascuda:
|
if utils.koboldai_vars.hascuda:
|
||||||
|
Reference in New Issue
Block a user