vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

ggml : drop support for QK_K=64 (#7473)

Browse source 
* ggml : drop support for QK_K=64

ggml-ci

* opencl : restore QK_K=256 define
This commit is contained in:
Georgi Gerganov 2024-05-23 10:00:21 +03:00 • committed by GitHub
parent 1b1e27cb49
commit e84b71c2c6
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
16 changed files with 26 additions and 4049 deletions
Download patch file Download diff file Expand all files Collapse all files

6
ggml-cuda/mmq.cu
View file

@ -826,11 +826,7 @@ template <int mmq_y, int nwarps, bool need_check> static __device__ __forceinlin
const block_q4_K * bxi = bx0 + i*blocks_per_row + kbxd;
#if QK_K == 256
x_dm[i * (WARP_SIZE/QI4_K) + i / QI4_K + kbxd] = bxi->dm;
#else
x_dm[i * (WARP_SIZE/QI4_K) + i / QI4_K + kbxd] = {bxi->dm[0], bxi->dm[1]};
#endif
}
#pragma unroll
@ -933,9 +929,7 @@ template <int mmq_y, int nwarps, bool need_check> static __device__ __forceinlin
const block_q5_K * bxi = bx0 + i*blocks_per_row + kbxd;
#if QK_K == 256
x_dm[i * (WARP_SIZE/QI5_K) + i / QI5_K + kbxd] = bxi->dm;
#endif
}
#pragma unroll