Skip to content

Implement customizable RoPE #2054

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 10 commits into from
Jul 15, 2023
Merged

Implement customizable RoPE #2054

Changes from 1 commit
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
dc0d0eb
Implement customizable RoPE
jxy Jun 30, 2023
1ae4318
ggml-metal: fix custom rope
jxy Jun 30, 2023
41819b0
common: fix argument names in help
jxy Jun 30, 2023
5c6eed3
llama: increase MEM_REQ_EVAL for MODEL_3B
jxy Jul 4, 2023
a728a0d
llama: make MEM_REQ_EVAL depend on n_ctx
jxy Jul 7, 2023
a3b4d93
server: use proper Content-Type in curl examples
jxy Jul 7, 2023
a6b5695
Merge remote-tracking branch 'upstream/master' into custom_rope
jxy Jul 13, 2023
da730c5
Merge branch 'custom_rope' of github.com:jxy/llama.cpp into custom_rope
jxy Jul 13, 2023
d0b6c94
style : minor fixes, mostly indentations
ggerganov Jul 15, 2023
6024bcc
ggml : fix asserts
ggerganov Jul 15, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions ggml.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15738,7 +15738,7 @@ static void ggml_compute_backward(struct ggml_context * ctx, struct ggml_tensor
// necessary for llama
if (src0->grad) {
assert(src1->type == GGML_TYPE_I32);
assert(ggml_nelements(src1) == 4);
assert(ggml_nelements(src1) == 6);
const int n_past = ((int32_t *) src1->data)[0];
const int n_dims = ((int32_t *) src1->data)[1];
const int mode = ((int32_t *) src1->data)[2];
Expand All @@ -15759,7 +15759,7 @@ static void ggml_compute_backward(struct ggml_context * ctx, struct ggml_tensor
{
if (src0->grad) {
assert(src1->type == GGML_TYPE_I32);
assert(ggml_nelements(src1) == 4);
assert(ggml_nelements(src1) == 3);
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't this be 6? Based on the code immediately after it should be at least 4, I think, not 3.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I went through the code an I also can't see why it's 3 when the lines just below it show it clearly taking 4 elements and looks like it designed to fail the assertion

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@ggerganov

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should be fixed in 513f861

const int n_past = ((int32_t *) src1->data)[0];
const int n_dims = ((int32_t *) src1->data)[1];
const int mode = ((int32_t *) src1->data)[2];
Expand Down