Skip to content

[None][perf] Explicitly set Kimi K3 router correction bias in FP32 #17061

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

Closed
jiaganc wants to merge 1 commit into from
+6 −6
Closed

[None][perf] Explicitly set Kimi K3 router correction bias in FP32 #17061

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
12 changes: 6 additions & 6 deletions tensorrt_llm/_torch/modules/kimi_k3_moe/kimi_k3_moe_gate.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,8 +29,7 @@
import torch
from torch import nn

from ..fused_moe.routing import (DeepSeekV3MoeRoutingMethod,
Deepseekv3RoutingImpl)
from ..fused_moe.routing import DeepSeekV3MoeRoutingMethod, Deepseekv3RoutingImpl


class KimiK3MoEGate(nn.Module):
Expand Down Expand Up @@ -89,7 +88,9 @@ def __init__(
self.weight = nn.Parameter(
torch.empty((self.num_experts, self.gating_dim), dtype=weight_dtype)
)
self.e_score_correction_bias = nn.Parameter(torch.empty(self.num_experts))
self.e_score_correction_bias = nn.Parameter(
torch.empty(self.num_experts, dtype=torch.float32)
)

self.softmax_routing_mutation = softmax_routing_mutation
self.biased_weights_mutation = biased_weights_mutation
Expand Down Expand Up @@ -145,8 +146,7 @@ def compute_logits(self, hidden_states: torch.Tensor) -> torch.Tensor:
parameter mapping identity.
"""
hidden_2d = hidden_states.reshape(-1, self.gating_dim)
if (self.weight.dtype == torch.bfloat16
and hidden_2d.dtype == torch.bfloat16):
if self.weight.dtype == torch.bfloat16 and hidden_2d.dtype == torch.bfloat16:
# Single bf16xbf16 -> fp32 GEMM (fp32 accumulate); no input
# upcast kernel, no fp32 splitK-reduce. K3's 896 experts miss
# the op's specialized 256-expert kernels and take its cublas
Expand Down Expand Up @@ -204,7 +204,7 @@ def forward(self, hidden_states: torch.Tensor) -> Tuple[torch.Tensor, torch.Tens
# indices (as ``torch.topk`` yields) and fp32 weights -- so every
# downstream consumer is byte-for-byte unaffected by the swap.
topk_weight, topk_idx = self._routing_impl.noaux_tc(
logits, self.e_score_correction_bias.float()
logits, self.e_score_correction_bias
)
return topk_idx.to(torch.int64), topk_weight.to(torch.float32)

Expand Down
Loading