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

fix baby llama, test-grad0

Browse source
This commit is contained in:
slaren 2023-07-24 16:38:37 +02:00
parent 24e53a1466
commit 8d7cfb42b7
2 changed files with 12 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files

20
examples/baby-llama/baby-llama.cpp
View file

@ -8,6 +8,8 @@
#pragma warning(disable: 4244 4267) // possible loss of data #pragma warning(disable: 4244 4267) // possible loss of data
#endif #endif
static const float rms_norm_eps = 1e-6f;
float frand() { float frand() {
return (float)rand()/(float)RAND_MAX; return (float)rand()/(float)RAND_MAX;
} }
@ -562,7 +564,7 @@ struct ggml_tensor * forward(
// norm // norm
{ {
// cur shape [n_embd,N,1,1] // cur shape [n_embd,N,1,1]
cur = ggml_rms_norm(ctx0, inpL); cur = ggml_rms_norm(ctx0, inpL, rms_norm_eps);
// cur = attention_norm*cur // cur = attention_norm*cur
cur = ggml_mul(ctx0, cur = ggml_mul(ctx0,
@ -685,7 +687,7 @@ struct ggml_tensor * forward(
// norm // norm
{ {
// cur shape [n_embd,N,1,1] // cur shape [n_embd,N,1,1]
cur = ggml_rms_norm(ctx0, inpFF); cur = ggml_rms_norm(ctx0, inpFF, rms_norm_eps);
// cur = ffn_norm*cur // cur = ffn_norm*cur
// cur shape [n_embd,N,1,1] // cur shape [n_embd,N,1,1]
@ -729,7 +731,7 @@ struct ggml_tensor * forward(
{ {
// inpL shape [n_embd,N,1,1] // inpL shape [n_embd,N,1,1]
inpL = ggml_rms_norm(ctx0, inpL); inpL = ggml_rms_norm(ctx0, inpL, rms_norm_eps);
// inpL = norm*inpL // inpL = norm*inpL
// inpL shape [n_embd,N,1,1] // inpL shape [n_embd,N,1,1]
@ -817,7 +819,7 @@ struct ggml_tensor * forward_batch(
// norm // norm
{ {
// cur shape [n_embd,N*n_batch,1,1] // cur shape [n_embd,N*n_batch,1,1]
cur = ggml_rms_norm(ctx0, inpL); cur = ggml_rms_norm(ctx0, inpL, rms_norm_eps);
assert_shape_2d(cur, n_embd, N*n_batch); assert_shape_2d(cur, n_embd, N*n_batch);
// cur = attention_norm*cur // cur = attention_norm*cur
@ -981,7 +983,7 @@ struct ggml_tensor * forward_batch(
// norm // norm
{ {
// cur shape [n_embd,N*n_batch,1,1] // cur shape [n_embd,N*n_batch,1,1]
cur = ggml_rms_norm(ctx0, inpFF); cur = ggml_rms_norm(ctx0, inpFF, rms_norm_eps);
assert_shape_2d(cur, n_embd, N*n_batch); assert_shape_2d(cur, n_embd, N*n_batch);
// cur = ffn_norm*cur // cur = ffn_norm*cur
@ -1034,7 +1036,7 @@ struct ggml_tensor * forward_batch(
{ {
// inpL shape [n_embd,N*n_batch,1,1] // inpL shape [n_embd,N*n_batch,1,1]
inpL = ggml_rms_norm(ctx0, inpL); inpL = ggml_rms_norm(ctx0, inpL, rms_norm_eps);
assert_shape_2d(inpL, n_embd, N*n_batch); assert_shape_2d(inpL, n_embd, N*n_batch);
// inpL = norm*inpL // inpL = norm*inpL
@ -1104,7 +1106,7 @@ struct ggml_tensor * forward_lora(
// norm // norm
{ {
// cur shape [n_embd,N,1,1] // cur shape [n_embd,N,1,1]
cur = ggml_rms_norm(ctx0, inpL); cur = ggml_rms_norm(ctx0, inpL, rms_norm_eps);
// cur = attention_norm*cur // cur = attention_norm*cur
cur = ggml_mul(ctx0, cur = ggml_mul(ctx0,
@ -1251,7 +1253,7 @@ struct ggml_tensor * forward_lora(
// norm // norm
{ {
// cur shape [n_embd,N,1,1] // cur shape [n_embd,N,1,1]
cur = ggml_rms_norm(ctx0, inpFF); cur = ggml_rms_norm(ctx0, inpFF, rms_norm_eps);
// cur = ffn_norm*cur // cur = ffn_norm*cur
// cur shape [n_embd,N,1,1] // cur shape [n_embd,N,1,1]
@ -1295,7 +1297,7 @@ struct ggml_tensor * forward_lora(
{ {
// inpL shape [n_embd,N,1,1] // inpL shape [n_embd,N,1,1]
inpL = ggml_rms_norm(ctx0, inpL); inpL = ggml_rms_norm(ctx0, inpL, rms_norm_eps);
// inpL = norm*inpL // inpL = norm*inpL
// inpL shape [n_embd,N,1,1] // inpL shape [n_embd,N,1,1]

2
tests/test-grad0.c
View file

@ -850,7 +850,7 @@ int main(int argc, const char ** argv) {
ggml_set_param(ctx0, x[i]); ggml_set_param(ctx0, x[i]);
} }
struct ggml_tensor * f = ggml_sum(ctx0, ggml_rms_norm(ctx0, x[0])); struct ggml_tensor * f = ggml_sum(ctx0, ggml_rms_norm(ctx0, x[0], 1e-6f));
check_gradient("rms_norm", ctx0, x, f, ndims, nargs, 1e-4f, 1.0f, INFINITY); check_gradient("rms_norm", ctx0, x, f, ndims, nargs, 1e-4f, 1.0f, INFINITY);
} }