Skip to content

[None][fix] Replace hash method with unique_id for cutedsl MoE runners. #9569

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
hyukn merged 1 commit into from
Dec 1, 2025
Merged

[None][fix] Replace hash method with unique_id for cutedsl MoE runners. #9569

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
51 changes: 46 additions & 5 deletions tensorrt_llm/_torch/custom_ops/cute_dsl_custom_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,7 +54,6 @@ def __init__(self, alpha: float, output_dtype: torch.dtype):
f"SM version {get_sm_version()} is not supported for {self.__class__.__name__}, it only supports SM 100"
)

# rewrite the hash function because the value of self.alpha doesn't affect the tactic.
def unique_id(self):
return (self.output_dtype, )

Expand Down Expand Up @@ -531,6 +530,17 @@ def __init__(self,
f"SM version {get_sm_version()} is not supported for {self.__class__.__name__}, it only supports SM 100"
)

def unique_id(self):
return (
self.num_experts,
self.top_k,
self.num_local_experts,
self.local_expert_offset,
self.tile_size,
self.output_dtype,
self.scaling_vector_size,
)

def get_valid_tactics(
self,
inputs: List[torch.Tensor],
Expand Down Expand Up @@ -571,7 +581,7 @@ def get_valid_tactics(
return valid_tactics

def get_tuning_config(self) -> TuningConfig:
key = hash(self)
key = self.unique_id()
if key not in self.__class__.tuning_config_cache:
helper = GroupedGemmInputsHelper(self.num_experts, self.top_k,
self.num_local_experts,
Expand Down Expand Up @@ -807,6 +817,17 @@ def __init__(self,
f"SM version {get_sm_version()} is not supported for {self.__class__.__name__}, it only supports SM 100"
)

def unique_id(self):
return (
self.num_experts,
self.top_k,
self.num_local_experts,
self.local_expert_offset,
self.tile_size,
self.output_dtype,
self.scaling_vector_size,
)

def get_valid_tactics(
self,
inputs: List[torch.Tensor],
Expand Down Expand Up @@ -847,7 +868,7 @@ def get_valid_tactics(
return valid_tactics

def get_tuning_config(self) -> TuningConfig:
key = hash(self)
key = self.unique_id()
if key not in self.__class__.tuning_config_cache:
helper = GroupedGemmInputsHelper(self.num_experts, self.top_k,
self.num_local_experts,
Expand Down Expand Up @@ -1124,6 +1145,16 @@ def __init__(self,
f"SM version {get_sm_version()} is not supported for {self.__class__.__name__}, it only supports SM 100"
)

def unique_id(self):
return (
self.num_experts,
self.top_k,
self.num_local_experts,
self.local_expert_offset,
self.tile_size,
self.scaling_vector_size,
)

def get_valid_tactics(
self,
inputs: List[torch.Tensor],
Expand Down Expand Up @@ -1164,7 +1195,7 @@ def get_valid_tactics(
return valid_tactics

def get_tuning_config(self) -> TuningConfig:
key = hash(self)
key = self.unique_id()
if key not in self.__class__.tuning_config_cache:
helper = GroupedGemmInputsHelper(self.num_experts, self.top_k,
self.num_local_experts,
Expand Down Expand Up @@ -1443,6 +1474,16 @@ def __init__(self,
self.output_dtype = output_dtype
self.scaling_vector_size = scaling_vector_size

def unique_id(self):
return (
self.num_experts,
self.top_k,
self.num_local_experts,
self.local_expert_offset,
self.output_dtype,
self.scaling_vector_size,
)

def get_valid_tactics(
self,
inputs: List[torch.Tensor],
Expand All @@ -1452,7 +1493,7 @@ def get_valid_tactics(
return [128]

def get_tuning_config(self) -> TuningConfig:
key = hash(self)
key = self.unique_id()
if key not in self.__class__.tuning_config_cache:
helper = FusedMoEInputsHelper(self.num_experts, self.top_k,
self.num_local_experts,
Expand Down
Loading