EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

CUDA: fix crash on large batch size for quant. MoE (#13537)

Browse Source
This commit is contained in:
Johannes Gäßler
2025-05-14 16:41:02 +02:00
committed by GitHub
parent b7d2672082
commit 4696d56749
2 changed files with 9 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/ggml-cuda/mmq.cu
View File

@@ -122,6 +122,7 @@ void ggml_cuda_mul_mat_q(
const int64_t s13 = src1->nb[3] / ts_src1;
quantize_mmq_q8_1_cuda(src1_d, nullptr, src1_q8_1.get(), src0->type,
ne10, s11, s12, s13, ne10_padded, ne11, ne12, ne13, stream);
CUDA_CHECK(cudaGetLastError());
}
const int64_t s12 = ne11*ne10_padded * sizeof(block_q8_1)/(QK8_1*sizeof(int));
@@ -205,6 +206,7 @@ void ggml_cuda_mul_mat_q(
const int64_t s13 = src1->nb[2] / ts_src1;
quantize_mmq_q8_1_cuda(src1_d, ids_src1_dev, src1_q8_1.get(), src0->type,
ne10, s11, s12, s13, ne10_padded, ne11_flat, ne12_flat, ne13_flat, stream);
CUDA_CHECK(cudaGetLastError());
}
const int64_t s12 = ne11*ne10_padded * sizeof(block_q8_1)/(QK8_1*sizeof(int));