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

fix: use correct Nvfp4Linear field names (fp4, scale_b, gsb)

Browse Source
This commit is contained in:
biondizzle
2026-06-01 06:47:15 +00:00
parent 16af96380f
commit 982f245c67
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tests/unit/test_fused_router.py
View File

@@ -38,9 +38,9 @@ def test_fused_router():
# The Nvfp4Linear expects stacked/contiguous weight tensors
from dsv4.layers.linear import Nvfp4Linear
gate_lin = Nvfp4Linear(in_features=K, out_features=N, device=device)
gate_lin._weight_fp4 = [w_fp4.contiguous()]
gate_lin._scale_b = [w_sf.contiguous()]
gate_lin._gsb = [ws2_val]
gate_lin.fp4 = [w_fp4.contiguous()]
gate_lin.scale_b = [w_sf.contiguous()]
gate_lin.gsb = [ws2_val]
gate_lin._activation_global_scale = None # set at runtime
gate_lin._ensure_stacked = lambda *a, **kw: None
gate_lin._ensure_initialized = lambda *a, **kw: None