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

Use ggml_norm not ggml_rms_norm

Browse source
This commit is contained in:
Galunid 2023-10-22 09:57:52 +02:00
parent 4fbce39089
commit a71041a05f
1 changed files with 12 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

19
llama.cpp
View file

@ -5797,7 +5797,7 @@ static struct ggml_cgraph * llm_build_stablelm(
const float freq_base = cparams.rope_freq_base;
const float freq_scale = cparams.rope_freq_scale;
const float norm_rms_eps = hparams.f_norm_rms_eps;
const float norm_eps = hparams.f_norm_eps;
const int n_gpu_layers = model.n_gpu_layers;
@ -5953,9 +5953,9 @@ static struct ggml_cgraph * llm_build_stablelm(
// norm
{
cur = ggml_rms_norm(ctx0, inpL, norm_rms_eps);
cur = ggml_norm(ctx0, inpL, norm_eps);
offload_func(cur);
ggml_set_name(cur, "rms_norm_0");
ggml_set_name(cur, "norm_0");
// cur = cur*attn_norm(broadcasted)
cur = ggml_mul(ctx0, cur, model.layers[il].attn_norm);
@ -6096,9 +6096,9 @@ static struct ggml_cgraph * llm_build_stablelm(
{
// norm
{
cur = ggml_rms_norm(ctx0, inpFF, norm_rms_eps);
cur = ggml_norm(ctx0, inpFF, norm_eps);
offload_func(cur);
ggml_set_name(cur, "rms_norm_1");
ggml_set_name(cur, "norm_1");
// cur = cur*ffn_norm(broadcasted)
cur = ggml_mul(ctx0, cur, model.layers[il].ffn_norm);
@ -6152,14 +6152,19 @@ static struct ggml_cgraph * llm_build_stablelm(
// norm
{
cur = ggml_rms_norm(ctx0, cur, norm_rms_eps);
cur = ggml_norm(ctx0, cur, norm_eps);
offload_func_nr(cur);
ggml_set_name(cur, "rms_norm_2");
ggml_set_name(cur, "norm_2");
// cur = cur*norm(broadcasted)
cur = ggml_mul(ctx0, cur, model.output_norm);
// offload_func_nr(cur); // TODO CPU + GPU mirrored backend
// ggml_set_name(cur, "result_norm");
cur = ggml_add(ctx0, cur, model.output_norm_b);
offload_func_nr(cur);
ggml_set_name(cur, "result_norm");
}
// lm_head