[Dependency] Update Humming and remove NVFP4 scale workaround (#61556)
Signed-off-by: jinzhen.ljz <jinzhen.ljz@antgroup.com> Co-authored-by: Codex <codex@openai.com> Co-authored-by: Misha Goin <mgoin64@gmail.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com>
This commit is contained in:
co-authored by
Codex
Misha Goin
Kevin H. Luu
parent
2e8a5930f6
commit
fe98d90a9d
@@ -9,6 +9,7 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
|
||||
kInt8DynamicTokenSym,
|
||||
)
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.utils.humming import create_humming_locks
|
||||
from vllm.utils.import_utils import has_humming
|
||||
|
||||
from .MPLinearKernel import MPLinearKernel, MPLinearLayerConfig
|
||||
@@ -71,7 +72,7 @@ class HummingLinearKernel(MPLinearKernel):
|
||||
input_schema=input_schema,
|
||||
)
|
||||
self.compute_config = get_humming_linear_compute_config()
|
||||
self.locks = torch.zeros(1024, dtype=torch.int32, device=layer.weight.device)
|
||||
self.locks = create_humming_locks(layer.weight.device)
|
||||
|
||||
def apply_weights(
|
||||
self,
|
||||
|
||||
Reference in New Issue
Block a user