frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 23:03:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[ROCm][Quantization][Kernel] Use FP8 FNUZ when OCP flag is 0 or undefined (#13851)

Browse Source 
Signed-off-by: Hollow Man <hollowman@opensuse.org>
This commit is contained in:
ℍ𝕠𝕝𝕝𝕠𝕨 𝕄𝕒𝕟
2025-02-27 04:39:10 +02:00
committed by GitHub
parent f95903909f
commit a31614e386
1 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
csrc/quantization/fp8/amd/quant_utils.cuh
View File

@ -24,12 +24,12 @@ __inline__ __device__ Tout scaled_vec_conversion(const Tin& x,
return x; return x;
} }
#if HIP_FP8_TYPE_FNUZ #if HIP_FP8_TYPE_OCP
using fp8_type = __hip_fp8_e4m3_fnuz;
using fp8x2_type = __hip_fp8x2_e4m3_fnuz;
#elif HIP_FP8_TYPE_OCP
using fp8_type = __hip_fp8_e4m3; using fp8_type = __hip_fp8_e4m3;
using fp8x2_type = __hip_fp8x2_e4m3; using fp8x2_type = __hip_fp8x2_e4m3;
#else
using fp8_type = __hip_fp8_e4m3_fnuz;
using fp8x2_type = __hip_fp8x2_e4m3_fnuz;
#endif #endif
// fp8 -> half // fp8 -> half