simcop2387/AutoGPTQ
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages 1 Wiki Activity

fix bug

Browse source
This commit is contained in:
qwopqwop200 2023-05-02 12:00:50 +09:00 • committed by GitHub
parent 144bd80436
commit 3c108d4232
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 64 additions and 60 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
quant_cuda/quant_cuda_kernel.cu
View file

@ -188,6 +188,7 @@ __global__ void VecQuant2MatMulKernel(
res += weight[k] * blockvec[k];
}
atomicAdd(&mul[b * width + w], res);
__syncthreads();
}
}
@ -332,6 +333,7 @@ __global__ void VecQuant3MatMulKernel(
res += weight[k] * blockvec[k];
}
atomicAdd(&mul[b * width + w], res);
__syncthreads();
}
}
@ -419,6 +421,7 @@ __global__ void VecQuant4MatMulKernel(
res += weight[k] * blockvec[k];
}
atomicAdd(&mul[b * width + w], res);
__syncthreads();
}
}
@ -505,5 +508,6 @@ __global__ void VecQuant8MatMulKernel(
res += weight[k] * blockvec[k];
}
atomicAdd(&mul[b * width + w], res);
__syncthreads();
}
}