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

ggml : ggml_rope now takes a vector with positions instead of n_past

Browse source
This commit is contained in:
Georgi Gerganov 2023-09-17 21:12:51 +03:00
parent 3b4bab6a38
commit 1fb033fd85
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
9 changed files with 270 additions and 131 deletions
Download patch file Download diff file Expand all files Collapse all files

17
ggml.h
View file

@ -1219,14 +1219,15 @@ extern "C" {
struct ggml_tensor * b);
// rotary position embedding
// if mode & 1 == 1, skip n_past elements
// if mode & 1 == 1, skip n_past elements (DEPRECATED)
// if mode & 2 == 1, GPT-NeoX style
// if mode & 4 == 1, ChatGLM style
// TODO: avoid creating a new tensor every time
//
// b is an int32 vector with size a->ne[2], it contains the positions
GGML_API struct ggml_tensor * ggml_rope(
struct ggml_context * ctx,
struct ggml_tensor * a,
int n_past,
struct ggml_tensor * b,
int n_dims,
int mode,
int n_ctx);
@ -1235,7 +1236,7 @@ extern "C" {
GGML_API struct ggml_tensor * ggml_rope_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a,
int n_past,
struct ggml_tensor * b,
int n_dims,
int mode,
int n_ctx);
@ -1244,7 +1245,7 @@ extern "C" {
GGML_API struct ggml_tensor * ggml_rope_custom(
struct ggml_context * ctx,
struct ggml_tensor * a,
int n_past,
struct ggml_tensor * b,
int n_dims,
int mode,
int n_ctx,
@ -1255,7 +1256,7 @@ extern "C" {
GGML_API struct ggml_tensor * ggml_rope_custom_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a,
int n_past,
struct ggml_tensor * b,
int n_dims,
int mode,
int n_ctx,
@ -1266,7 +1267,7 @@ extern "C" {
GGML_API struct ggml_tensor * ggml_rope_xpos_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a,
int n_past,
struct ggml_tensor * b,
int n_dims,
float base,
bool down);
@ -1276,7 +1277,7 @@ extern "C" {
GGML_API struct ggml_tensor * ggml_rope_back(
struct ggml_context * ctx,
struct ggml_tensor * a,
int n_past,
struct ggml_tensor * b,
int n_dims,
int mode,
int n_ctx,