mirror of
https://github.com/dogkeeper886/ollama37.git
synced 2025-12-13 01:07:12 +00:00
llama: update vendored code to commit 46e3556 (#8308)
This commit is contained in:
Jeffrey Morgan
297
llama/ggml.c
vendored
297
llama/ggml.c
vendored
@@ -1,5 +1,5 @@
|
||||
/**
|
||||
* llama.cpp - commit ba1cb19cdd0d92e012e0f6e009e0620f854b6afd - do not edit this file
|
||||
* llama.cpp - commit 46e3556e01b824e52395fb050b29804b6cff2a7c - do not edit this file
|
||||
*
|
||||
* MIT License
|
||||
*
|
||||
@@ -3788,104 +3788,10 @@ struct ggml_tensor * ggml_clamp(
|
||||
return result;
|
||||
}
|
||||
|
||||
// ggml_conv_1d
|
||||
|
||||
static int64_t ggml_calc_conv_output_size(int64_t ins, int64_t ks, int s, int p, int d) {
|
||||
return (ins + 2 * p - d * (ks - 1) - 1) / s + 1;
|
||||
}
|
||||
|
||||
GGML_API struct ggml_tensor * ggml_conv_1d(
|
||||
struct ggml_context * ctx,
|
||||
struct ggml_tensor * a,
|
||||
struct ggml_tensor * b,
|
||||
int s0,
|
||||
int p0,
|
||||
int d0) {
|
||||
struct ggml_tensor * im2col = ggml_im2col(ctx, a, b, s0, 0, p0, 0, d0, 0, false, GGML_TYPE_F16); // [N, OL, IC * K]
|
||||
|
||||
struct ggml_tensor * result =
|
||||
ggml_mul_mat(ctx,
|
||||
ggml_reshape_2d(ctx, im2col, im2col->ne[0], (im2col->ne[2] * im2col->ne[1])), // [N, OL, IC * K] => [N*OL, IC * K]
|
||||
ggml_reshape_2d(ctx, a, (a->ne[0] * a->ne[1]), a->ne[2])); // [OC，IC, K] => [OC, IC * K]
|
||||
|
||||
result = ggml_reshape_3d(ctx, result, im2col->ne[1], a->ne[2], im2col->ne[2]); // [N, OC, OL]
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
// ggml_conv_1d_ph
|
||||
|
||||
struct ggml_tensor* ggml_conv_1d_ph(
|
||||
struct ggml_context * ctx,
|
||||
struct ggml_tensor * a,
|
||||
struct ggml_tensor * b,
|
||||
int s,
|
||||
int d) {
|
||||
return ggml_conv_1d(ctx, a, b, s, a->ne[0] / 2, d);
|
||||
}
|
||||
|
||||
// ggml_conv_transpose_1d
|
||||
|
||||
static int64_t ggml_calc_conv_transpose_1d_output_size(int64_t ins, int64_t ks, int s, int p, int d) {
|
||||
return (ins - 1) * s - 2 * p + d * (ks - 1) + 1;
|
||||
}
|
||||
|
||||
GGML_API struct ggml_tensor * ggml_conv_transpose_1d(
|
||||
struct ggml_context * ctx,
|
||||
struct ggml_tensor * a,
|
||||
struct ggml_tensor * b,
|
||||
int s0,
|
||||
int p0,
|
||||
int d0) {
|
||||
GGML_ASSERT(ggml_is_matrix(b));
|
||||
GGML_ASSERT(a->ne[2] == b->ne[1]);
|
||||
GGML_ASSERT(a->ne[3] == 1);
|
||||
|
||||
GGML_ASSERT(p0 == 0);
|
||||
GGML_ASSERT(d0 == 1);
|
||||
|
||||
const int64_t ne[4] = {
|
||||
ggml_calc_conv_transpose_1d_output_size(b->ne[0], a->ne[0], s0, 0 /*p0*/, 1 /*d0*/),
|
||||
a->ne[1], b->ne[2], 1,
|
||||
};
|
||||
struct ggml_tensor * result = ggml_new_tensor(ctx, GGML_TYPE_F32, 4, ne);
|
||||
|
||||
int32_t params[] = { s0, p0, d0 };
|
||||
ggml_set_op_params(result, params, sizeof(params));
|
||||
|
||||
result->op = GGML_OP_CONV_TRANSPOSE_1D;
|
||||
result->src[0] = a;
|
||||
result->src[1] = b;
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
// ggml_conv_depthwise
|
||||
|
||||
struct ggml_tensor * ggml_conv_depthwise_2d(
|
||||
struct ggml_context * ctx,
|
||||
struct ggml_tensor * a,
|
||||
struct ggml_tensor * b,
|
||||
int s0,
|
||||
int s1,
|
||||
int p0,
|
||||
int p1,
|
||||
int d0,
|
||||
int d1) {
|
||||
struct ggml_tensor * new_a = ggml_reshape_4d(ctx, a, a->ne[0], a->ne[1], 1, a->ne[2] * a->ne[3]);
|
||||
struct ggml_tensor * im2col = ggml_im2col(ctx, new_a,
|
||||
ggml_reshape_4d(ctx, b, b->ne[0], b->ne[1], 1, b->ne[2] * b->ne[3]),
|
||||
s0, s1, p0, p1, d0, d1, true, GGML_TYPE_F16); // [N * IC, OH, OW, KH * KW]
|
||||
struct ggml_tensor * new_b = ggml_reshape_4d(ctx, im2col, im2col->ne[0], im2col->ne[2] * im2col->ne[1], b->ne[2], b->ne[3]); // [N * IC, OH, OW, KH * KW] => [N, IC, OH * OW, KH * KW]
|
||||
|
||||
new_a = ggml_reshape_4d(ctx, new_a, (new_a->ne[0] * new_a->ne[1]), new_a->ne[2], new_a->ne[3], 1); // [OC，1, KH, KW] => [1, OC, 1, KH * KW]
|
||||
struct ggml_tensor * result = ggml_mul_mat(ctx, new_a, new_b);
|
||||
result = ggml_reshape_4d(ctx, result, im2col->ne[1], im2col->ne[2], b->ne[2], b->ne[3]); // [N, OC, OH, OW]
|
||||
|
||||
return result;
|
||||
}
|
||||
// ggml_conv_2d
|
||||
|
||||
// im2col: [N, IC, IH, IW] => [N, OH, OW, IC*KH*KW]
|
||||
// a: [OC，IC, KH, KW]
|
||||
// b: [N, IC, IH, IW]
|
||||
@@ -3902,10 +3808,11 @@ struct ggml_tensor * ggml_im2col(
|
||||
int d1,
|
||||
bool is_2D,
|
||||
enum ggml_type dst_type) {
|
||||
if(is_2D) {
|
||||
if (is_2D) {
|
||||
GGML_ASSERT(a->ne[2] == b->ne[2]);
|
||||
} else {
|
||||
GGML_ASSERT(a->ne[1] == b->ne[1]);
|
||||
//GGML_ASSERT(b->ne[1] % a->ne[1] == 0);
|
||||
GGML_ASSERT(b->ne[1] == a->ne[1]);
|
||||
GGML_ASSERT(b->ne[3] == 1);
|
||||
}
|
||||
|
||||
@@ -3956,6 +3863,108 @@ struct ggml_tensor * ggml_im2col_back(
|
||||
return result;
|
||||
}
|
||||
|
||||
// ggml_conv_1d
|
||||
|
||||
struct ggml_tensor * ggml_conv_1d(
|
||||
struct ggml_context * ctx,
|
||||
struct ggml_tensor * a,
|
||||
struct ggml_tensor * b,
|
||||
int s0,
|
||||
int p0,
|
||||
int d0) {
|
||||
struct ggml_tensor * im2col = ggml_im2col(ctx, a, b, s0, 0, p0, 0, d0, 0, false, GGML_TYPE_F16); // [N, OL, IC * K]
|
||||
|
||||
struct ggml_tensor * result =
|
||||
ggml_mul_mat(ctx,
|
||||
ggml_reshape_2d(ctx, im2col, im2col->ne[0], (im2col->ne[2] * im2col->ne[1])), // [N, OL, IC * K] => [N*OL, IC * K]
|
||||
ggml_reshape_2d(ctx, a, (a->ne[0] * a->ne[1]), a->ne[2])); // [OC，IC, K] => [OC, IC * K]
|
||||
|
||||
result = ggml_reshape_3d(ctx, result, im2col->ne[1], a->ne[2], im2col->ne[2]); // [N, OC, OL]
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
// ggml_conv_1d_ph
|
||||
|
||||
struct ggml_tensor* ggml_conv_1d_ph(
|
||||
struct ggml_context * ctx,
|
||||
struct ggml_tensor * a,
|
||||
struct ggml_tensor * b,
|
||||
int s,
|
||||
int d) {
|
||||
return ggml_conv_1d(ctx, a, b, s, a->ne[0] / 2, d);
|
||||
}
|
||||
|
||||
// ggml_conv_1d_dw
|
||||
|
||||
struct ggml_tensor * ggml_conv_1d_dw(
|
||||
struct ggml_context * ctx,
|
||||
struct ggml_tensor * a,
|
||||
struct ggml_tensor * b,
|
||||
int s0,
|
||||
int p0,
|
||||
int d0) {
|
||||
struct ggml_tensor * new_a = ggml_reshape_4d(ctx, a, a->ne[0], 1, a->ne[1], a->ne[2]);
|
||||
struct ggml_tensor * new_b = ggml_reshape_4d(ctx, b, b->ne[0], 1, b->ne[1], b->ne[2]);
|
||||
|
||||
struct ggml_tensor * im2col = ggml_im2col(ctx, new_a, new_b, s0, 0, p0, 0, d0, 0, false, GGML_TYPE_F16);
|
||||
|
||||
struct ggml_tensor * result = ggml_mul_mat(ctx, im2col, a);
|
||||
|
||||
result = ggml_reshape_3d(ctx, result, b->ne[0], b->ne[1], 1);
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
// ggml_conv_1d_dw_ph
|
||||
|
||||
struct ggml_tensor * ggml_conv_1d_dw_ph(
|
||||
struct ggml_context * ctx,
|
||||
struct ggml_tensor * a,
|
||||
struct ggml_tensor * b,
|
||||
int s0,
|
||||
int d0) {
|
||||
return ggml_conv_1d_dw(ctx, a, b, s0, a->ne[0] / 2, d0);
|
||||
}
|
||||
|
||||
// ggml_conv_transpose_1d
|
||||
|
||||
static int64_t ggml_calc_conv_transpose_1d_output_size(int64_t ins, int64_t ks, int s, int p, int d) {
|
||||
return (ins - 1) * s - 2 * p + d * (ks - 1) + 1;
|
||||
}
|
||||
|
||||
GGML_API struct ggml_tensor * ggml_conv_transpose_1d(
|
||||
struct ggml_context * ctx,
|
||||
struct ggml_tensor * a,
|
||||
struct ggml_tensor * b,
|
||||
int s0,
|
||||
int p0,
|
||||
int d0) {
|
||||
GGML_ASSERT(ggml_is_matrix(b));
|
||||
GGML_ASSERT(a->ne[2] == b->ne[1]);
|
||||
GGML_ASSERT(a->ne[3] == 1);
|
||||
|
||||
GGML_ASSERT(p0 == 0);
|
||||
GGML_ASSERT(d0 == 1);
|
||||
|
||||
const int64_t ne[4] = {
|
||||
ggml_calc_conv_transpose_1d_output_size(b->ne[0], a->ne[0], s0, 0 /*p0*/, 1 /*d0*/),
|
||||
a->ne[1], b->ne[2], 1,
|
||||
};
|
||||
struct ggml_tensor * result = ggml_new_tensor(ctx, GGML_TYPE_F32, 4, ne);
|
||||
|
||||
int32_t params[] = { s0, p0, d0 };
|
||||
ggml_set_op_params(result, params, sizeof(params));
|
||||
|
||||
result->op = GGML_OP_CONV_TRANSPOSE_1D;
|
||||
result->src[0] = a;
|
||||
result->src[1] = b;
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
// ggml_conv_2d
|
||||
|
||||
// a: [OC，IC, KH, KW]
|
||||
// b: [N, IC, IH, IW]
|
||||
// result: [N, OC, OH, OW]
|
||||
@@ -4001,6 +4010,31 @@ struct ggml_tensor * ggml_conv_2d_s1_ph(
|
||||
return ggml_conv_2d(ctx, a, b, 1, 1, a->ne[0] / 2, a->ne[1] / 2, 1, 1);
|
||||
}
|
||||
|
||||
// ggml_conv_2d_dw
|
||||
|
||||
struct ggml_tensor * ggml_conv_2d_dw(
|
||||
struct ggml_context * ctx,
|
||||
struct ggml_tensor * a,
|
||||
struct ggml_tensor * b,
|
||||
int s0,
|
||||
int s1,
|
||||
int p0,
|
||||
int p1,
|
||||
int d0,
|
||||
int d1) {
|
||||
struct ggml_tensor * new_a = ggml_reshape_4d(ctx, a, a->ne[0], a->ne[1], 1, a->ne[2] * a->ne[3]);
|
||||
struct ggml_tensor * im2col = ggml_im2col(ctx, new_a,
|
||||
ggml_reshape_4d(ctx, b, b->ne[0], b->ne[1], 1, b->ne[2] * b->ne[3]),
|
||||
s0, s1, p0, p1, d0, d1, true, GGML_TYPE_F16); // [N * IC, OH, OW, KH * KW]
|
||||
struct ggml_tensor * new_b = ggml_reshape_4d(ctx, im2col, im2col->ne[0], im2col->ne[2] * im2col->ne[1], b->ne[2], b->ne[3]); // [N * IC, OH, OW, KH * KW] => [N, IC, OH * OW, KH * KW]
|
||||
|
||||
new_a = ggml_reshape_4d(ctx, new_a, (new_a->ne[0] * new_a->ne[1]), new_a->ne[2], new_a->ne[3], 1); // [OC，1, KH, KW] => [1, OC, 1, KH * KW]
|
||||
struct ggml_tensor * result = ggml_mul_mat(ctx, new_a, new_b);
|
||||
result = ggml_reshape_4d(ctx, result, im2col->ne[1], im2col->ne[2], b->ne[2], b->ne[3]); // [N, OC, OH, OW]
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
// ggml_conv_transpose_2d_p0
|
||||
|
||||
static int64_t ggml_calc_conv_transpose_output_size(int64_t ins, int64_t ks, int s, int p) {
|
||||
@@ -6084,12 +6118,12 @@ struct ggml_tensor * ggml_graph_get_tensor(const struct ggml_cgraph * cgraph, co
|
||||
|
||||
struct ggml_tensor * ggml_graph_get_grad(const struct ggml_cgraph * cgraph, const struct ggml_tensor * node) {
|
||||
const size_t igrad = ggml_hash_find(&cgraph->visited_hash_set, node);
|
||||
return igrad != GGML_HASHSET_FULL && ggml_bitset_get(cgraph->visited_hash_set.used, igrad) ? cgraph->grads[igrad] : NULL;
|
||||
return igrad != GGML_HASHSET_FULL && ggml_bitset_get(cgraph->visited_hash_set.used, igrad) && cgraph->grads ? cgraph->grads[igrad] : NULL;
|
||||
}
|
||||
|
||||
struct ggml_tensor * ggml_graph_get_grad_acc(const struct ggml_cgraph * cgraph, const struct ggml_tensor * node) {
|
||||
const size_t igrad = ggml_hash_find(&cgraph->visited_hash_set, node);
|
||||
return igrad != GGML_HASHSET_FULL && ggml_bitset_get(cgraph->visited_hash_set.used, igrad) ? cgraph->grad_accs[igrad] : NULL;
|
||||
return igrad != GGML_HASHSET_FULL && ggml_bitset_get(cgraph->visited_hash_set.used, igrad) && cgraph->grad_accs ? cgraph->grad_accs[igrad] : NULL;
|
||||
}
|
||||
|
||||
void ggml_graph_print(const struct ggml_cgraph * cgraph) {
|
||||
@@ -6536,7 +6570,7 @@ struct gguf_context {
|
||||
void * data;
|
||||
};
|
||||
|
||||
static size_t gguf_type_size(enum gguf_type type) {
|
||||
size_t gguf_type_size(enum gguf_type type) {
|
||||
GGML_ASSERT(0 <= type && type < GGUF_TYPE_COUNT);
|
||||
return GGUF_TYPE_SIZE[type];
|
||||
}
|
||||
@@ -6664,13 +6698,7 @@ struct gguf_context * gguf_init_empty(void) {
|
||||
return ctx;
|
||||
}
|
||||
|
||||
struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_params params) {
|
||||
FILE * file = ggml_fopen(fname, "rb");
|
||||
if (!file) {
|
||||
fprintf(stderr, "%s: failed to open '%s': '%s'\n", __func__, fname, strerror(errno));
|
||||
return NULL;
|
||||
}
|
||||
|
||||
struct gguf_context * gguf_init_from_file_impl(FILE * file, struct gguf_init_params params) {
|
||||
// offset from start of file
|
||||
size_t offset = 0;
|
||||
|
||||
@@ -6683,7 +6711,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
for (uint32_t i = 0; i < sizeof(magic); i++) {
|
||||
if (magic[i] != GGUF_MAGIC[i]) {
|
||||
fprintf(stderr, "%s: invalid magic characters '%c%c%c%c'\n", __func__, magic[0], magic[1], magic[2], magic[3]);
|
||||
fclose(file);
|
||||
return NULL;
|
||||
}
|
||||
}
|
||||
@@ -6694,7 +6721,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
struct gguf_context * ctx = calloc(1, sizeof(struct gguf_context));
|
||||
if (!ctx) {
|
||||
fprintf(stderr, "%s: failed to allocate memory for context\n", __func__);
|
||||
fclose(file);
|
||||
return NULL;
|
||||
}
|
||||
|
||||
@@ -6712,7 +6738,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
|
||||
if (ctx->header.version == 1) {
|
||||
fprintf(stderr, "%s: GGUFv1 is no longer supported. please use a more up-to-date version\n", __func__);
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
}
|
||||
@@ -6725,7 +6750,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
|
||||
if (!ok) {
|
||||
fprintf(stderr, "%s: failed to read header\n", __func__);
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
}
|
||||
@@ -6735,12 +6759,13 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
{
|
||||
const uint64_t n_kv = ctx->header.n_kv;
|
||||
|
||||
ctx->kv = calloc(n_kv, sizeof(struct gguf_kv));
|
||||
if (!ctx->kv) {
|
||||
fprintf(stderr, "%s: failed to allocate memory for kv pairs\n", __func__);
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
if (n_kv > 0) {
|
||||
ctx->kv = calloc(n_kv, sizeof(struct gguf_kv));
|
||||
if (!ctx->kv) {
|
||||
fprintf(stderr, "%s: failed to allocate memory for kv pairs\n", __func__);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
}
|
||||
}
|
||||
|
||||
for (uint64_t i = 0; i < n_kv; ++i) {
|
||||
@@ -6787,7 +6812,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
// prevent from integer overflow in the malloc below
|
||||
if (kv->value.arr.n >= SIZE_MAX/gguf_type_size(kv->value.arr.type)) {
|
||||
fprintf(stderr, "%s: array size is too large (%" PRIu64 ")\n", __func__, kv->value.arr.n);
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
}
|
||||
@@ -6795,7 +6819,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
kv->value.arr.data = calloc(kv->value.arr.n, gguf_type_size(kv->value.arr.type));
|
||||
if (!kv->value.arr.data) {
|
||||
fprintf(stderr, "%s: failed to allocate memory for array\n", __func__);
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
}
|
||||
@@ -6807,7 +6830,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
// prevent from integer overflow in the malloc below
|
||||
if (kv->value.arr.n >= SIZE_MAX/sizeof(struct gguf_str)) {
|
||||
fprintf(stderr, "%s: array size is too large (%" PRIu64 ")\n", __func__, kv->value.arr.n);
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
}
|
||||
@@ -6815,7 +6837,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
kv->value.arr.data = calloc(kv->value.arr.n, sizeof(struct gguf_str));
|
||||
if (!kv->value.arr.data) {
|
||||
fprintf(stderr, "%s: failed to allocate memory for array\n", __func__);
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
}
|
||||
@@ -6846,7 +6867,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
|
||||
if (!ok) {
|
||||
fprintf(stderr, "%s: failed to read key-value pairs\n", __func__);
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
}
|
||||
@@ -6857,7 +6877,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
ctx->infos = calloc(ctx->header.n_tensors, sizeof(struct gguf_tensor_info));
|
||||
if (!ctx->infos) {
|
||||
fprintf(stderr, "%s: failed to allocate memory for tensor infos\n", __func__);
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
}
|
||||
@@ -6893,7 +6912,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
|
||||
if (!ok) {
|
||||
fprintf(stderr, "%s: failed to read tensor info\n", __func__);
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
}
|
||||
@@ -6936,7 +6954,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
// this tensor type support have been removed:
|
||||
fprintf(stderr, "%s: tensor '%s' of type %d: %s\n",
|
||||
__func__, info->name.data, (int) info->type, ggml_type_name(info->type));
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
}
|
||||
@@ -6944,7 +6961,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
if (ne % ggml_blck_size(info->type) != 0) {
|
||||
fprintf(stderr, "%s: tensor '%s' of type %d (%s) number of elements (%" PRId64 ") is not a multiple of block size (%" PRId64 ")\n",
|
||||
__func__, info->name.data, (int) info->type, ggml_type_name(info->type), ne, ggml_blck_size(info->type));
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
}
|
||||
@@ -6976,7 +6992,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
*params.ctx = ggml_init(pdata);
|
||||
if (*params.ctx == NULL) {
|
||||
fprintf(stderr, "%s: failed to initialize context\n", __func__);
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
}
|
||||
@@ -6995,7 +7010,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
|
||||
if (!ok) {
|
||||
fprintf(stderr, "%s: failed to read tensor data\n", __func__);
|
||||
fclose(file);
|
||||
ggml_free(ctx_data);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
@@ -7034,7 +7048,6 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
|
||||
if (!ok) {
|
||||
fprintf(stderr, "%s: failed to read the tensor data\n", __func__);
|
||||
fclose(file);
|
||||
ggml_free(ctx_data);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
@@ -7043,11 +7056,21 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
||||
ggml_set_no_alloc(ctx_data, params.no_alloc);
|
||||
}
|
||||
|
||||
fclose(file);
|
||||
|
||||
return ctx;
|
||||
}
|
||||
|
||||
struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_params params) {
|
||||
FILE * file = ggml_fopen(fname, "rb");
|
||||
if (!file) {
|
||||
fprintf(stderr, "%s: failed to open '%s': '%s'\n", __func__, fname, strerror(errno));
|
||||
return NULL;
|
||||
}
|
||||
|
||||
struct gguf_context * result = gguf_init_from_file_impl(file, params);
|
||||
fclose(file);
|
||||
return result;
|
||||
}
|
||||
|
||||
void gguf_free(struct gguf_context * ctx) {
|
||||
if (ctx == NULL) {
|
||||
return;
|
||||
@@ -7507,13 +7530,7 @@ void gguf_set_tensor_data(struct gguf_context * ctx, const char * name, const vo
|
||||
// fwrite(val, sizeof(char), size, file);
|
||||
//}
|
||||
|
||||
struct gguf_buf {
|
||||
void * data;
|
||||
size_t size;
|
||||
size_t offset;
|
||||
};
|
||||
|
||||
static struct gguf_buf gguf_buf_init(size_t size) {
|
||||
struct gguf_buf gguf_buf_init(size_t size) {
|
||||
struct gguf_buf buf = {
|
||||
/*buf.data =*/ size == 0 ? NULL : GGML_CALLOC(1, size),
|
||||
/*buf.size =*/ size,
|
||||
@@ -7523,7 +7540,7 @@ static struct gguf_buf gguf_buf_init(size_t size) {
|
||||
return buf;
|
||||
}
|
||||
|
||||
static void gguf_buf_free(struct gguf_buf buf) {
|
||||
void gguf_buf_free(struct gguf_buf buf) {
|
||||
if (buf.data) {
|
||||
GGML_FREE(buf.data);
|
||||
}
|
||||
@@ -7561,7 +7578,7 @@ static void gguf_bwrite_el(struct gguf_buf * buf, const void * val, size_t el_si
|
||||
buf->offset += el_size;
|
||||
}
|
||||
|
||||
static void gguf_write_to_buf(const struct gguf_context * ctx, struct gguf_buf * buf, bool only_meta) {
|
||||
void gguf_write_to_buf(const struct gguf_context * ctx, struct gguf_buf * buf, bool only_meta) {
|
||||
// write header
|
||||
gguf_bwrite_el(buf, &ctx->header.magic, sizeof(ctx->header.magic));
|
||||
gguf_bwrite_el(buf, &ctx->header.version, sizeof(ctx->header.version));
|
||||
|
||||
Reference in New Issue
Block a user