xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-07 23:02:15 +08:00
Code Issues Packages Projects Releases Wiki Activity

Missing comment explaining VDR variable in GGUF kernels (#13290)

Browse Source
This commit is contained in:
Szymon Ożóg 2025-02-21 07:06:54 +01:00 committed by GitHub
parent 31aa045c11
commit 1cdc88614a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
csrc/quantization/gguf/vecdotq.cuh
View File

@ -37,6 +37,8 @@ static __device__ __forceinline__ int get_int_from_uint8_aligned(const uint8_t *
return *((const int *) (x8 + sizeof(int) * i32)); // assume at least 4 byte alignment return *((const int *) (x8 + sizeof(int) * i32)); // assume at least 4 byte alignment
} }
// VDR = vec dot ratio, how many contiguous integers each thread processes when the vec dot kernel is called
// MMVQ = mul_mat_vec_q, MMQ = mul_mat_q
#define VDR_Q4_0_Q8_1_MMVQ 2 #define VDR_Q4_0_Q8_1_MMVQ 2
#define VDR_Q4_0_Q8_1_MMQ 4 #define VDR_Q4_0_Q8_1_MMQ 4