fix: add missing import for quantize_nvfp4_gpu in linear.py fixed-gsa path

This commit is contained in:
2026-06-02 04:28:29 +00:00
parent f52eedbdce
commit 7e3fb5f4d0

View File

@@ -176,6 +176,7 @@ class Nvfp4Linear:
x_fp4, x_sf, gsa_gpu = quantize_nvfp4_gpu_fused(hidden_states)
self._gsa_buf.copy_(gsa_gpu[:1].reshape(1)) # GPU → GPU, no sync
else:
from dsv4.ops.quantize import quantize_nvfp4_gpu
self._gsa_buf.fill_(self._activation_global_scale)
x_fp4, x_sf = quantize_nvfp4_gpu(hidden_states, self._activation_global_scale)