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

bug fix for scale backward pass

Browse source 
use sum instead of mean for gradient of scalar scale parameter
This commit is contained in:
xaedes 2023-04-25 22:25:53 +02:00
parent 671e5922e2
commit a367eb9eda
No known key found for this signature in database
GPG key ID: 30030EDD817EA2B1
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml.c
View file

@ -12802,7 +12802,7 @@ static void ggml_compute_backward(struct ggml_context * ctx, struct ggml_tensor
src1->grad = src1->grad =
ggml_add_impl(ctx, ggml_add_impl(ctx,
src1->grad, src1->grad,
ggml_mean(ctx, ggml_mul_impl(ctx, tensor->grad, src0, false)), ggml_sum(ctx, ggml_mul_impl(ctx, tensor->grad, src0, false)),
inplace); inplace);
} }
} break; } break;