aditya/llama.cpp
1
0
Fork 0
mirror of https://git.adityakumar.xyz/llama.cpp.git synced 2024-11-09 23:29:44 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

ggml : change ne to int64_t (#626)

Browse source
This commit is contained in:
Marian Cepok 2023-04-02 12:21:31 +02:00 committed by GitHub
parent 6e7801d08d
commit c0bb1d3ce2
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 375 additions and 374 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

691
ggml.c
View file

File diff suppressed because it is too large Load diff

52
ggml.h
View file

@ -258,11 +258,11 @@ struct ggml_tensor {
enum ggml_type type; enum ggml_type type;
int n_dims; int n_dims;
int ne[GGML_MAX_DIMS]; // number of elements int64_t ne[GGML_MAX_DIMS]; // number of elements
size_t nb[GGML_MAX_DIMS]; // stride in bytes: size_t nb[GGML_MAX_DIMS]; // stride in bytes:
// nb[0] = sizeof(type) // nb[0] = sizeof(type)
// nb[1] = nb[0] * ne[0] + padding // nb[1] = nb[0] * ne[0] + padding
// nb[i] = nb[i-1] * ne[i-1] // nb[i] = nb[i-1] * ne[i-1]
// compute data // compute data
enum ggml_op op; enum ggml_op op;
@ -328,8 +328,8 @@ int64_t ggml_cycles_per_ms(void);
void ggml_print_object (const struct ggml_object * obj); void ggml_print_object (const struct ggml_object * obj);
void ggml_print_objects(const struct ggml_context * ctx); void ggml_print_objects(const struct ggml_context * ctx);
int ggml_nelements(const struct ggml_tensor * tensor); int64_t ggml_nelements(const struct ggml_tensor * tensor);
size_t ggml_nbytes (const struct ggml_tensor * tensor); size_t ggml_nbytes (const struct ggml_tensor * tensor);
int ggml_blck_size (enum ggml_type type); int ggml_blck_size (enum ggml_type type);
size_t ggml_type_size (enum ggml_type type); // size in bytes for all elements in a block size_t ggml_type_size (enum ggml_type type); // size in bytes for all elements in a block
@ -355,33 +355,33 @@ struct ggml_tensor * ggml_new_tensor(
struct ggml_context * ctx, struct ggml_context * ctx,
enum ggml_type type, enum ggml_type type,
int n_dims, int n_dims,
const int *ne); const int64_t *ne);
struct ggml_tensor * ggml_new_tensor_1d( struct ggml_tensor * ggml_new_tensor_1d(
struct ggml_context * ctx, struct ggml_context * ctx,
enum ggml_type type, enum ggml_type type,
int ne0); int64_t ne0);
struct ggml_tensor * ggml_new_tensor_2d( struct ggml_tensor * ggml_new_tensor_2d(
struct ggml_context * ctx, struct ggml_context * ctx,
enum ggml_type type, enum ggml_type type,
int ne0, int64_t ne0,
int ne1); int64_t ne1);
struct ggml_tensor * ggml_new_tensor_3d( struct ggml_tensor * ggml_new_tensor_3d(
struct ggml_context * ctx, struct ggml_context * ctx,
enum ggml_type type, enum ggml_type type,
int ne0, int64_t ne0,
int ne1, int64_t ne1,
int ne2); int64_t ne2);
struct ggml_tensor * ggml_new_tensor_4d( struct ggml_tensor * ggml_new_tensor_4d(
struct ggml_context * ctx, struct ggml_context * ctx,
enum ggml_type type, enum ggml_type type,
int ne0, int64_t ne0,
int ne1, int64_t ne1,
int ne2, int64_t ne2,
int ne3); int64_t ne3);
struct ggml_tensor * ggml_new_i32(struct ggml_context * ctx, int32_t value); struct ggml_tensor * ggml_new_i32(struct ggml_context * ctx, int32_t value);
struct ggml_tensor * ggml_new_f32(struct ggml_context * ctx, float value); struct ggml_tensor * ggml_new_f32(struct ggml_context * ctx, float value);
@ -531,30 +531,30 @@ struct ggml_tensor * ggml_reshape(
struct ggml_tensor * ggml_reshape_2d( struct ggml_tensor * ggml_reshape_2d(
struct ggml_context * ctx, struct ggml_context * ctx,
struct ggml_tensor * a, struct ggml_tensor * a,
int ne0, int64_t ne0,
int ne1); int64_t ne1);
// return view(a) // return view(a)
// TODO: when we start computing gradient, make a copy instead of view // TODO: when we start computing gradient, make a copy instead of view
struct ggml_tensor * ggml_reshape_3d( struct ggml_tensor * ggml_reshape_3d(
struct ggml_context * ctx, struct ggml_context * ctx,
struct ggml_tensor * a, struct ggml_tensor * a,
int ne0, int64_t ne0,
int ne1, int64_t ne1,
int ne2); int64_t ne2);
// offset in bytes // offset in bytes
struct ggml_tensor * ggml_view_1d( struct ggml_tensor * ggml_view_1d(
struct ggml_context * ctx, struct ggml_context * ctx,
struct ggml_tensor * a, struct ggml_tensor * a,
int ne0, int64_t ne0,
size_t offset); size_t offset);
struct ggml_tensor * ggml_view_2d( struct ggml_tensor * ggml_view_2d(
struct ggml_context * ctx, struct ggml_context * ctx,
struct ggml_tensor * a, struct ggml_tensor * a,
int ne0, int64_t ne0,
int ne1, int64_t ne1,
size_t nb1, // row stride in bytes size_t nb1, // row stride in bytes
size_t offset); size_t offset);

6
llama.cpp
View file

@ -256,8 +256,8 @@ static bool kv_cache_init(
const int n_embd = hparams.n_embd; const int n_embd = hparams.n_embd;
const int n_layer = hparams.n_layer; const int n_layer = hparams.n_layer;
const int n_mem = n_layer*n_ctx; const int64_t n_mem = (int64_t)n_layer*n_ctx;
const int n_elements = n_embd*n_mem; const int64_t n_elements = n_embd*n_mem;
cache.buf.resize(2u*n_elements*ggml_type_size(wtype) + 2u*MB); cache.buf.resize(2u*n_elements*ggml_type_size(wtype) + 2u*MB);
@ -679,7 +679,7 @@ static bool llama_model_load(
return false; return false;
} }
if (tensor->ne[0] != ne[0] || tensor->ne[1] != ne[1]) { if (tensor->ne[0] != ne[0] || tensor->ne[1] != ne[1]) {
fprintf(stderr, "%s: tensor '%s' has wrong shape in model file: got [%d, %d], expected [%d, %d]\n", fprintf(stderr, "%s: tensor '%s' has wrong shape in model file: got [%" PRId64 ", %" PRId64 "], expected [%d, %d]\n",
__func__, name.data(), tensor->ne[0], tensor->ne[1], ne[0], ne[1]); __func__, name.data(), tensor->ne[0], tensor->ne[1], ne[0], ne[1]);
return false; return false;
} }