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

Merge branch 'master' into gg/flash-attn

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2024-04-29 17:19:25 +03:00
commit a1616e9f72
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
82 changed files with 3896 additions and 1063 deletions
Download patch file Download diff file Expand all files Collapse all files

43
ggml.c
View file

@ -20959,7 +20959,7 @@ static void gguf_free_kv(struct gguf_kv * kv) {
}
struct gguf_context * gguf_init_empty(void) {
struct gguf_context * ctx = GGML_ALIGNED_MALLOC(sizeof(struct gguf_context));
struct gguf_context * ctx = GGML_CALLOC(1, sizeof(struct gguf_context));
memcpy(ctx->header.magic, GGUF_MAGIC, sizeof(ctx->header.magic));
ctx->header.version = GGUF_VERSION;
@ -21004,7 +21004,7 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
bool ok = true;
struct gguf_context * ctx = GGML_ALIGNED_MALLOC(sizeof(struct gguf_context));
struct gguf_context * ctx = GGML_CALLOC(1, sizeof(struct gguf_context));
// read the header
{
@ -21041,9 +21041,13 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
// read the kv pairs
{
ctx->kv = GGML_MALLOC(ctx->header.n_kv * sizeof(struct gguf_kv));
const uint64_t n_kv = ctx->header.n_kv;
for (uint64_t i = 0; i < ctx->header.n_kv; ++i) {
// header.n_kv will hold the actual value of pairs that were successfully read in the loop below
ctx->header.n_kv = 0;
ctx->kv = GGML_CALLOC(n_kv, sizeof(struct gguf_kv));
for (uint64_t i = 0; i < n_kv; ++i) {
struct gguf_kv * kv = &ctx->kv[i];
//fprintf(stderr, "%s: reading kv %d\n", __func__, i);
@ -21092,7 +21096,7 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
return NULL;
}
kv->value.arr.data = GGML_MALLOC(kv->value.arr.n * gguf_type_size(kv->value.arr.type));
kv->value.arr.data = GGML_CALLOC(kv->value.arr.n, gguf_type_size(kv->value.arr.type));
ok = ok && gguf_fread_el(file, kv->value.arr.data, kv->value.arr.n * gguf_type_size(kv->value.arr.type), &offset);
} break;
@ -21106,7 +21110,7 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
return NULL;
}
kv->value.arr.data = GGML_MALLOC(kv->value.arr.n * sizeof(struct gguf_str));
kv->value.arr.data = GGML_CALLOC(kv->value.arr.n, sizeof(struct gguf_str));
for (uint64_t j = 0; j < kv->value.arr.n; ++j) {
ok = ok && gguf_fread_str(file, &((struct gguf_str *) kv->value.arr.data)[j], &offset);
@ -21122,6 +21126,8 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
if (!ok) {
break;
}
ctx->header.n_kv++;
}
if (!ok) {
@ -21134,7 +21140,7 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
// read the tensor infos
{
ctx->infos = GGML_MALLOC(ctx->header.n_tensors * sizeof(struct gguf_tensor_info));
ctx->infos = GGML_CALLOC(ctx->header.n_tensors, sizeof(struct gguf_tensor_info));
for (uint64_t i = 0; i < ctx->header.n_tensors; ++i) {
struct gguf_tensor_info * info = &ctx->infos[i];
@ -21155,8 +21161,17 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
ok = ok && gguf_fread_el (file, &info->type, sizeof(info->type), &offset);
ok = ok && gguf_fread_el (file, &info->offset, sizeof(info->offset), &offset);
// TODO: return an error instead of crashing with GGML_ASSERT
gguf_tensor_info_sanitize(info);
// make sure there is no duplicated tensor names
for (uint64_t j = 0; j < i; ++j) {
if (strcmp(info->name.data, ctx->infos[j].name.data) == 0) {
fprintf(stderr, "%s: duplicated tensor name %s\n", __func__, info->name.data);
ok = false;
}
}
if (!ok) {
fprintf(stderr, "%s: failed to read tensor info\n", __func__);
fclose(file);
@ -21325,7 +21340,7 @@ void gguf_free(struct gguf_context * ctx) {
GGML_FREE(ctx->infos);
}
GGML_ALIGNED_FREE(ctx);
GGML_FREE(ctx);
}
const char * gguf_type_name(enum gguf_type type) {
@ -21636,7 +21651,7 @@ void gguf_set_arr_data(struct gguf_context * ctx, const char * key, enum gguf_ty
ctx->kv[idx].type = GGUF_TYPE_ARRAY;
ctx->kv[idx].value.arr.type = type;
ctx->kv[idx].value.arr.n = n;
ctx->kv[idx].value.arr.data = GGML_MALLOC(n*gguf_type_size(type));
ctx->kv[idx].value.arr.data = GGML_CALLOC(n, gguf_type_size(type));
memcpy(ctx->kv[idx].value.arr.data, data, n*gguf_type_size(type));
}
@ -21646,7 +21661,7 @@ void gguf_set_arr_str(struct gguf_context * ctx, const char * key, const char **
ctx->kv[idx].type = GGUF_TYPE_ARRAY;
ctx->kv[idx].value.arr.type = GGUF_TYPE_STRING;
ctx->kv[idx].value.arr.n = n;
ctx->kv[idx].value.arr.data = GGML_MALLOC(n*sizeof(struct gguf_str));
ctx->kv[idx].value.arr.data = GGML_CALLOC(n, sizeof(struct gguf_str));
for (int i = 0; i < n; i++) {
struct gguf_str * str = &((struct gguf_str *)ctx->kv[idx].value.arr.data)[i];
str->n = strlen(data[i]);
@ -21673,7 +21688,7 @@ void gguf_set_kv(struct gguf_context * ctx, struct gguf_context * src) {
case GGUF_TYPE_ARRAY:
{
if (src->kv[i].value.arr.type == GGUF_TYPE_STRING) {
const char ** data = GGML_MALLOC(src->kv[i].value.arr.n*sizeof(char *));
const char ** data = GGML_CALLOC(src->kv[i].value.arr.n, sizeof(char *));
for (uint32_t j = 0; j < src->kv[i].value.arr.n; j++) {
data[j] = ((struct gguf_str *)src->kv[i].value.arr.data)[j].data;
}
@ -21693,6 +21708,10 @@ void gguf_set_kv(struct gguf_context * ctx, struct gguf_context * src) {
void gguf_add_tensor(
struct gguf_context * ctx,
const struct ggml_tensor * tensor) {
if (gguf_find_tensor(ctx, tensor->name) != -1) {
GGML_ASSERT(false && "duplicated tensor name");
}
const int idx = ctx->header.n_tensors;
ctx->infos = realloc(ctx->infos, (idx + 1)*sizeof(struct gguf_tensor_info));
@ -21761,7 +21780,7 @@ struct gguf_buf {
static struct gguf_buf gguf_buf_init(size_t size) {
struct gguf_buf buf = {
/*buf.data =*/ size == 0 ? NULL : GGML_MALLOC(size),
/*buf.data =*/ size == 0 ? NULL : GGML_CALLOC(1, size),
/*buf.size =*/ size,
/*buf.offset =*/ 0,
};