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 17, 05:30 UTC
[ROCm] Fix build problem resulted from previous commit related to FP8 kv-cache support (#2790)
Hongxia Yang committed
Feb 7, 2024 at 01:36 UTC
c81dddb45c71e630b907f9d84686ecd73b4105c7
2 files changed
+16
Dockerfile.rocm
+1
@@ -76,6 +76,7 @@ RUN cd /app \
76
&& cd vllm \
77
&& pip install -U -r requirements-rocm.txt \
78
&& bash patch_xformers.rocm.sh \
79
+ && patch /opt/rocm/include/hip/amd_detail/amd_hip_bf16.h /app/vllm/rocm_patch/rocm_bf16.patch \
80
&& python3 setup.py install \
81
&& cd ..
82
rocm_patch/rocm_bf16.patch
new
+15
@@ -0,0 +1,15 @@
1
+--- amd_hip_bf16.h 2024-02-06 18:28:58.268699142 +0000
2
++++ amd_hip_bf16.h.new 2024-02-06 18:28:31.988647133 +0000
3
+@@ -90,10 +90,10 @@
4
+ #include "math_fwd.h" // ocml device functions
5
+
6
+ #if defined(__HIPCC_RTC__)
7
+-#define __HOST_DEVICE__ __device__
8
++#define __HOST_DEVICE__ __device__ static
9
+ #else
10
+ #include <climits>
11
+-#define __HOST_DEVICE__ __host__ __device__
12
++#define __HOST_DEVICE__ __host__ __device__ static inline
13
+ #endif
14
+
15
+ // Since we are using unsigned short to represent data in bfloat16, it can be of different sizes on