biondizzle/nvfp4-megamoe-kernel
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

chore: remove unused _expert_id_range after bincount migration

Browse Source
This commit is contained in:
biondizzle
2026-05-20 02:17:44 +00:00
parent 84a2f6d441
commit 4f178d6e9c
1 changed files with 0 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
cutedsl/runner.py
View File

@@ -72,7 +72,6 @@ class CuTeDSLMoERunner:
# Pre-allocated cudagraph buffers (set in _allocate_buffers)
self._token_indices = None
self._expert_id_range = None
self._expert_offsets_buf = None
self._per_expert_scale_bufs_l1 = None
self._per_expert_scale_bufs_l2 = None
@@ -260,9 +259,6 @@ class CuTeDSLMoERunner:
warmup_compilation(self.num_experts, K_packed, N_packed_l1, self.device)
warmup_compilation(self.num_experts, K_packed, N_packed_l2, self.device)
self._expert_id_range = torch.arange(
self.num_experts, dtype=torch.int32
).to(self.device)
self._expert_offsets_buf = torch.zeros(
self.num_experts + 1, dtype=torch.int32, device=self.device
)
@@ -392,7 +388,6 @@ class CuTeDSLMoERunner:
# Quantize slot_hidden for GEMM
slot_x_fp4, slot_x_sf = quantize_activation_nvfp4(slot_hidden, l1_gs)
expert_id_range = self._expert_id_range
tokens_per_expert = torch.bincount(sorted_ids, minlength=self.num_experts)[:self.num_experts].int()
expert_offsets = self._expert_offsets_buf
expert_offsets.zero_()
@@ -495,7 +490,6 @@ class CuTeDSLMoERunner:
sorted_token_ids = token_indices[sort_idx]
# Expert offsets (real token counts)
expert_id_range = self._expert_id_range
tokens_per_expert = torch.bincount(sorted_ids, minlength=self.num_experts)[:self.num_experts].int()
expert_offsets = self._expert_offsets_buf
expert_offsets.zero_()