A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 5, 16:00 UTC
[Compile] Fix nvfp4 compile warning (#38573)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Apr 1, 2026 at 14:28 UTC
c9a9db0e023a03f7e2c40e43be302fb07512946a
1 file changed
+7
-4
csrc/libtorch_stable/quantization/fp4/nvfp4_quant_entry.cu
+7
-4
@@ -58,16 +58,19 @@ void silu_and_mul_scaled_fp4_experts_quant_sm1xxa(
58
torch::stable::Tensor const& output_scale_offset_by_experts);
59
#endif
60
61
+#if (defined(ENABLE_NVFP4_SM100) && ENABLE_NVFP4_SM100) || \
62
+ (defined(ENABLE_NVFP4_SM120) && ENABLE_NVFP4_SM120)
63
static bool nvfp4_quant_sm_supported() {
64
const int32_t sm = get_sm_version_num();
63
-#if defined(ENABLE_NVFP4_SM100) && ENABLE_NVFP4_SM100
65
+ #if defined(ENABLE_NVFP4_SM100) && ENABLE_NVFP4_SM100
66
if (sm >= 100 && sm < 120) return true;
65
-#endif
66
-#if defined(ENABLE_NVFP4_SM120) && ENABLE_NVFP4_SM120
67
+ #endif
68
+ #if defined(ENABLE_NVFP4_SM120) && ENABLE_NVFP4_SM120
69
if (sm >= 120 && sm < 130) return true;
68
-#endif
70
+ #endif
71
return false;
72
}
73
+#endif
74
75
void scaled_fp4_quant_out(torch::stable::Tensor const& input,
76
torch::stable::Tensor const& input_sf,