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 14, 05:30 UTC
[Bugfix] Avoid uninitialized usage of azp_val when AZP is false. (#24335)
Signed-off-by: Mohan Kumar Kumar <mohan.cbein@gmail.com> Signed-off-by: mohankku <mohan.cbein@gmail.com>
mohankku committed
Sep 6, 2025 at 08:17 UTC
0eadaeff7e51ada11771b31d2056bda05b012f5c
1 file changed
+2
-1
csrc/cpu/dnnl_kernels.cpp
+2
-1
@@ -145,7 +145,8 @@ void dynamic_scaled_int8_quant_impl(const scalar_t* input, int8_t* output,
145
}
146
}
147
148
- float scale_val, azp_val;
148
+ float scale_val;
149
+ float azp_val = 0.0f;
150
if constexpr (AZP) {
151
float max_scalar = max_value.reduce_max();
152
float min_scalar = min_value.reduce_min();