CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-13 10:57:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into compilade/mamba2

Browse Source
This commit is contained in:
Francis Couture-Harpin
2025-02-26 15:32:20 -05:00
parent 1ee6c482d0 a800ae46da
commit c9ecf620d6
702 changed files with 98035 additions and 62199 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
ggml/include/ggml-amx.h
View File

@@ -1,25 +0,0 @@
#pragma once
#include "ggml.h"
#include "ggml-backend.h"
#ifdef __cplusplus
extern "C" {
#endif
// buffer_type API
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_amx_buffer_type(void);
GGML_BACKEND_API bool ggml_backend_is_amx(ggml_backend_t backend);
// backend API
GGML_BACKEND_API ggml_backend_t ggml_backend_amx_init(void);
GGML_BACKEND_API void ggml_backend_amx_set_n_threads(ggml_backend_t backend_amx, int n_threads);
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_amx_reg(void);
#ifdef __cplusplus
}
#endif

3
ggml/include/ggml-backend.h
View File

@@ -203,6 +203,8 @@ extern "C" {
// Backend registry
//
GGML_API void ggml_backend_device_register(ggml_backend_dev_t device);
// Backend (reg) enumeration
GGML_API size_t ggml_backend_reg_count(void);
GGML_API ggml_backend_reg_t ggml_backend_reg_get(size_t index);
@@ -228,6 +230,7 @@ extern "C" {
GGML_API void ggml_backend_unload(ggml_backend_reg_t reg);
// Load all known backends from dynamic libraries
GGML_API void ggml_backend_load_all(void);
GGML_API void ggml_backend_load_all_from_path(const char * dir_path);
//
// Backend scheduler

1
ggml/include/ggml-cpp.h
View File

@@ -7,6 +7,7 @@
#include "ggml.h"
#include "ggml-alloc.h"
#include "ggml-backend.h"
#include "gguf.h"
#include <memory>
// Smart pointers for ggml types

22
ggml/include/ggml-cpu.h
View File

@@ -8,7 +8,7 @@ extern "C" {
#endif
// the compute plan that needs to be prepared for ggml_graph_compute()
// since https://github.com/ggerganov/ggml/issues/287
// since https://github.com/ggml-org/ggml/issues/287
struct ggml_cplan {
size_t work_size; // size of work buffer, calculated by `ggml_graph_plan()`
uint8_t * work_data; // work buffer, to be allocated by caller before calling to `ggml_graph_compute()`
@@ -91,35 +91,28 @@ extern "C" {
GGML_BACKEND_API int ggml_cpu_has_neon (void);
GGML_BACKEND_API int ggml_cpu_has_arm_fma (void);
GGML_BACKEND_API int ggml_cpu_has_fp16_va (void);
GGML_BACKEND_API int ggml_cpu_has_dotprod (void);
GGML_BACKEND_API int ggml_cpu_has_matmul_int8(void);
GGML_BACKEND_API int ggml_cpu_has_sve (void);
GGML_BACKEND_API int ggml_cpu_get_sve_cnt (void); // sve vector length in bytes
GGML_BACKEND_API int ggml_cpu_has_sme (void);
// other
GGML_BACKEND_API int ggml_cpu_has_riscv_v (void);
GGML_BACKEND_API int ggml_cpu_has_vsx (void);
GGML_BACKEND_API int ggml_cpu_has_vxe (void);
GGML_BACKEND_API int ggml_cpu_has_wasm_simd (void);
GGML_BACKEND_API int ggml_cpu_has_llamafile (void);
// Internal types and functions exposed for tests and benchmarks
typedef void (*ggml_from_float_to_mat_t)
(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t nr, int64_t k, int64_t bs);
typedef void (*ggml_vec_dot_t) (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT x, size_t bx,
const void * GGML_RESTRICT y, size_t by, int nrc);
typedef void (*ggml_gemv_t) (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT x,
const void * GGML_RESTRICT y, int nr, int nc);
typedef void (*ggml_gemm_t) (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT x,
const void * GGML_RESTRICT y, int nr, int nc);
struct ggml_type_traits_cpu {
ggml_from_float_t from_float;
ggml_from_float_to_mat_t from_float_to_mat;
ggml_vec_dot_t vec_dot;
enum ggml_type vec_dot_type;
int64_t nrows; // number of rows to process simultaneously
int64_t ncols; // number of columns to process simultaneously
ggml_gemv_t gemv;
ggml_gemm_t gemm;
};
GGML_BACKEND_API const struct ggml_type_traits_cpu * ggml_get_type_traits_cpu(enum ggml_type type);
@@ -139,13 +132,6 @@ extern "C" {
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_cpu_reg(void);
#ifdef GGML_USE_CPU_HBM
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_cpu_hbm_buffer_type(void);
#endif
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_cpu_aarch64_buffer_type(void);
GGML_BACKEND_API bool ggml_backend_cpu_buft_is_aarch64(ggml_backend_buffer_type_t buft);
#ifdef __cplusplus
}
#endif

2
ggml/include/ggml-metal.h
View File

@@ -45,7 +45,7 @@ GGML_BACKEND_API bool ggml_backend_is_metal(ggml_backend_t backend);
GGML_DEPRECATED(
GGML_BACKEND_API ggml_backend_buffer_t ggml_backend_metal_buffer_from_ptr(void * data, size_t size, size_t max_size),
"obsoleted by the new device interface - https://github.com/ggerganov/llama.cpp/pull/9713");
"obsoleted by the new device interface - https://github.com/ggml-org/llama.cpp/pull/9713");
GGML_BACKEND_API void ggml_backend_metal_set_abort_callback(ggml_backend_t backend, ggml_abort_callback abort_callback, void * user_data);

26
ggml/include/ggml-opencl.h Normal file
View File

@@ -0,0 +1,26 @@
#ifndef GGML_OPENCL_H
#define GGML_OPENCL_H
#include "ggml.h"
#include "ggml-backend.h"
#ifdef __cplusplus
extern "C" {
#endif
//
// backend API
//
GGML_BACKEND_API ggml_backend_t ggml_backend_opencl_init(void);
GGML_BACKEND_API bool ggml_backend_is_opencl(ggml_backend_t backend);
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_opencl_buffer_type(void);
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_opencl_host_buffer_type(void);
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_opencl_reg(void);
#ifdef __cplusplus
}
#endif
#endif // GGML_OPENCL_H

2
ggml/include/ggml-vulkan.h
View File

@@ -10,8 +10,6 @@ extern "C" {
#define GGML_VK_NAME "Vulkan"
#define GGML_VK_MAX_DEVICES 16
GGML_BACKEND_API void ggml_vk_instance_init(void);
// backend API
GGML_BACKEND_API ggml_backend_t ggml_backend_vk_init(size_t dev_num);

284
ggml/include/ggml.h
View File

@@ -198,7 +198,7 @@
#ifndef __GNUC__
# define GGML_ATTRIBUTE_FORMAT(...)
#elif defined(__MINGW32__)
#elif defined(__MINGW32__) && !defined(__clang__)
# define GGML_ATTRIBUTE_FORMAT(...) __attribute__((format(gnu_printf, __VA_ARGS__)))
#else
# define GGML_ATTRIBUTE_FORMAT(...) __attribute__((format(printf, __VA_ARGS__)))
@@ -237,13 +237,9 @@
#define GGML_EXIT_SUCCESS 0
#define GGML_EXIT_ABORTED 1
#define GGML_ROPE_TYPE_NEOX 2
#define GGUF_MAGIC "GGUF"
#define GGUF_VERSION 3
#define GGUF_DEFAULT_ALIGNMENT 32
#define GGML_ROPE_TYPE_NEOX 2
#define GGML_ROPE_TYPE_MROPE 8
#define GGML_ROPE_TYPE_VISION 24
#define GGML_UNUSED(x) (void)(x)
@@ -384,12 +380,15 @@ extern "C" {
GGML_TYPE_F64 = 28,
GGML_TYPE_IQ1_M = 29,
GGML_TYPE_BF16 = 30,
GGML_TYPE_Q4_0_4_4 = 31,
GGML_TYPE_Q4_0_4_8 = 32,
GGML_TYPE_Q4_0_8_8 = 33,
// GGML_TYPE_Q4_0_4_4 = 31, support has been removed from gguf files
// GGML_TYPE_Q4_0_4_8 = 32,
// GGML_TYPE_Q4_0_8_8 = 33,
GGML_TYPE_TQ1_0 = 34,
GGML_TYPE_TQ2_0 = 35,
GGML_TYPE_COUNT,
// GGML_TYPE_IQ4_NL_4_4 = 36,
// GGML_TYPE_IQ4_NL_4_8 = 37,
// GGML_TYPE_IQ4_NL_8_8 = 38,
GGML_TYPE_COUNT = 39,
};
// precision
@@ -398,12 +397,6 @@ extern "C" {
GGML_PREC_F32,
};
enum ggml_backend_type {
GGML_BACKEND_TYPE_CPU = 0,
GGML_BACKEND_TYPE_GPU = 10,
GGML_BACKEND_TYPE_GPU_SPLIT = 20,
};
// model file types
enum ggml_ftype {
GGML_FTYPE_UNKNOWN = -1,
@@ -430,9 +423,6 @@ extern "C" {
GGML_FTYPE_MOSTLY_IQ4_XS = 22, // except 1d tensors
GGML_FTYPE_MOSTLY_IQ1_M = 23, // except 1d tensors
GGML_FTYPE_MOSTLY_BF16 = 24, // except 1d tensors
GGML_FTYPE_MOSTLY_Q4_0_4_4 = 25, // except 1d tensors
GGML_FTYPE_MOSTLY_Q4_0_4_8 = 26, // except 1d tensors
GGML_FTYPE_MOSTLY_Q4_0_8_8 = 27, // except 1d tensors
};
// available tensor operations:
@@ -496,6 +486,7 @@ extern "C" {
GGML_OP_POOL_2D_BACK,
GGML_OP_UPSCALE, // nearest interpolate
GGML_OP_PAD,
GGML_OP_PAD_REFLECT_1D,
GGML_OP_ARANGE,
GGML_OP_TIMESTEP_EMBEDDING,
GGML_OP_ARGSORT,
@@ -510,6 +501,7 @@ extern "C" {
GGML_OP_GET_REL_POS,
GGML_OP_ADD_REL_POS,
GGML_OP_RWKV_WKV6,
GGML_OP_GATED_LINEAR_ATTN,
GGML_OP_UNARY,
@@ -584,8 +576,6 @@ extern "C" {
struct ggml_tensor {
enum ggml_type type;
GGML_DEPRECATED(enum ggml_backend_type backend, "use the buffer type to find the storage location of the tensor");
struct ggml_backend_buffer * buffer;
int64_t ne[GGML_MAX_DIMS]; // number of elements
@@ -1394,16 +1384,20 @@ extern "C" {
float scale,
float max_bias);
GGML_API struct ggml_tensor * ggml_soft_max_back(
GGML_API struct ggml_tensor * ggml_soft_max_ext_back(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
struct ggml_tensor * b,
float scale,
float max_bias);
// in-place, returns view(a)
GGML_API struct ggml_tensor * ggml_soft_max_back_inplace(
GGML_API struct ggml_tensor * ggml_soft_max_ext_back_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
struct ggml_tensor * b,
float scale,
float max_bias);
// rotary position embedding
// if (mode & 1) - skip n_past elements (NOT SUPPORTED)
@@ -1442,6 +1436,22 @@ extern "C" {
float beta_fast,
float beta_slow);
GGML_API struct ggml_tensor * ggml_rope_multi(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
struct ggml_tensor * c,
int n_dims,
int sections[4],
int mode,
int n_ctx_orig,
float freq_base,
float freq_scale,
float ext_factor,
float attn_factor,
float beta_fast,
float beta_slow);
// in-place, returns view(a)
GGML_API struct ggml_tensor * ggml_rope_ext_inplace(
struct ggml_context * ctx,
@@ -1494,7 +1504,7 @@ extern "C" {
// rotary position embedding backward, i.e compute dx from dy
// a - dy
GGML_API struct ggml_tensor * ggml_rope_back(
GGML_API struct ggml_tensor * ggml_rope_ext_back(
struct ggml_context * ctx,
struct ggml_tensor * a, // gradients of ggml_rope result
struct ggml_tensor * b, // positions
@@ -1509,6 +1519,23 @@ extern "C" {
float beta_fast,
float beta_slow);
GGML_API struct ggml_tensor * ggml_rope_multi_back(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
struct ggml_tensor * c,
int n_dims,
int sections[4],
int mode,
int n_ctx_orig,
float freq_base,
float freq_scale,
float ext_factor,
float attn_factor,
float beta_fast,
float beta_slow);
// clamp
// in-place, returns view(a)
GGML_API struct ggml_tensor * ggml_clamp(
@@ -1545,17 +1572,6 @@ extern "C" {
int d1, // dilation dimension 1
bool is_2D);
GGML_API struct ggml_tensor * ggml_conv_depthwise_2d(
struct ggml_context * ctx,
struct ggml_tensor * a, // convolution kernel
struct ggml_tensor * b, // data
int s0, // stride dimension 0
int s1, // stride dimension 1
int p0, // padding dimension 0
int p1, // padding dimension 1
int d0, // dilation dimension 0
int d1); // dilation dimension 1
GGML_API struct ggml_tensor * ggml_conv_1d(
struct ggml_context * ctx,
struct ggml_tensor * a, // convolution kernel
@@ -1573,6 +1589,23 @@ extern "C" {
int s, // stride
int d); // dilation
// depthwise
// TODO: this is very likely wrong for some cases! - needs more testing
GGML_API struct ggml_tensor * ggml_conv_1d_dw(
struct ggml_context * ctx,
struct ggml_tensor * a, // convolution kernel
struct ggml_tensor * b, // data
int s0, // stride
int p0, // padding
int d0); // dilation
GGML_API struct ggml_tensor * ggml_conv_1d_dw_ph(
struct ggml_context * ctx,
struct ggml_tensor * a, // convolution kernel
struct ggml_tensor * b, // data
int s0, // stride
int d0); // dilation
GGML_API struct ggml_tensor * ggml_conv_transpose_1d(
struct ggml_context * ctx,
struct ggml_tensor * a, // convolution kernel
@@ -1592,7 +1625,6 @@ extern "C" {
int d0, // dilation dimension 0
int d1); // dilation dimension 1
// kernel size is a->ne[0] x a->ne[1]
// stride is equal to kernel size
// padding is zero
@@ -1619,6 +1651,18 @@ extern "C" {
struct ggml_tensor * a,
struct ggml_tensor * b);
// depthwise
GGML_API struct ggml_tensor * ggml_conv_2d_dw(
struct ggml_context * ctx,
struct ggml_tensor * a, // convolution kernel
struct ggml_tensor * b, // data
int s0, // stride dimension 0
int s1, // stride dimension 1
int p0, // padding dimension 0
int p1, // padding dimension 1
int d0, // dilation dimension 0
int d1); // dilation dimension 1
GGML_API struct ggml_tensor * ggml_conv_transpose_2d_p0(
struct ggml_context * ctx,
struct ggml_tensor * a,
@@ -1692,6 +1736,13 @@ extern "C" {
int p2,
int p3);
// pad each dimension with reflection: [a, b, c, d] -> [b, a, b, c, d, c]
GGML_API struct ggml_tensor * ggml_pad_reflect_1d(
struct ggml_context * ctx,
struct ggml_tensor * a,
int p0,
int p1);
// Ref: https://github.com/CompVis/stable-diffusion/blob/main/ldm/modules/diffusionmodules/util.py#L151
// timesteps: [N,]
// return: [N, dim]
@@ -1724,7 +1775,7 @@ extern "C" {
struct ggml_tensor * a,
int k);
#define GGML_KQ_MASK_PAD 32
#define GGML_KQ_MASK_PAD 64
// q: [n_embd, n_batch, n_head, 1]
// k: [n_embd, n_kv, n_head_kv, 1]
@@ -1831,6 +1882,15 @@ extern "C" {
struct ggml_tensor * td,
struct ggml_tensor * state);
GGML_API struct ggml_tensor * ggml_gated_linear_attn(
struct ggml_context * ctx,
struct ggml_tensor * k,
struct ggml_tensor * v,
struct ggml_tensor * q,
struct ggml_tensor * g,
struct ggml_tensor * state,
float scale);
// custom operators
typedef void (*ggml_unary_op_f32_t) (const int, float *, const float *);
@@ -2069,137 +2129,19 @@ extern "C" {
int64_t n_per_row,
const float * imatrix);
//
// gguf
//
enum gguf_type {
GGUF_TYPE_UINT8 = 0,
GGUF_TYPE_INT8 = 1,
GGUF_TYPE_UINT16 = 2,
GGUF_TYPE_INT16 = 3,
GGUF_TYPE_UINT32 = 4,
GGUF_TYPE_INT32 = 5,
GGUF_TYPE_FLOAT32 = 6,
GGUF_TYPE_BOOL = 7,
GGUF_TYPE_STRING = 8,
GGUF_TYPE_ARRAY = 9,
GGUF_TYPE_UINT64 = 10,
GGUF_TYPE_INT64 = 11,
GGUF_TYPE_FLOAT64 = 12,
GGUF_TYPE_COUNT, // marks the end of the enum
};
struct gguf_context;
struct gguf_init_params {
bool no_alloc;
// if not NULL, create a ggml_context and allocate the tensor data in it
struct ggml_context ** ctx;
};
GGML_API struct gguf_context * gguf_init_empty(void);
GGML_API struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_params params);
//GGML_API struct gguf_context * gguf_init_from_buffer(..);
GGML_API void gguf_free(struct gguf_context * ctx);
GGML_API const char * gguf_type_name(enum gguf_type type);
GGML_API int gguf_get_version (const struct gguf_context * ctx);
GGML_API size_t gguf_get_alignment (const struct gguf_context * ctx);
GGML_API size_t gguf_get_data_offset(const struct gguf_context * ctx);
GGML_API void * gguf_get_data (const struct gguf_context * ctx);
GGML_API int gguf_get_n_kv(const struct gguf_context * ctx);
GGML_API int gguf_find_key(const struct gguf_context * ctx, const char * key);
GGML_API const char * gguf_get_key (const struct gguf_context * ctx, int key_id);
GGML_API enum gguf_type gguf_get_kv_type (const struct gguf_context * ctx, int key_id);
GGML_API enum gguf_type gguf_get_arr_type(const struct gguf_context * ctx, int key_id);
// will abort if the wrong type is used for the key
GGML_API uint8_t gguf_get_val_u8 (const struct gguf_context * ctx, int key_id);
GGML_API int8_t gguf_get_val_i8 (const struct gguf_context * ctx, int key_id);
GGML_API uint16_t gguf_get_val_u16 (const struct gguf_context * ctx, int key_id);
GGML_API int16_t gguf_get_val_i16 (const struct gguf_context * ctx, int key_id);
GGML_API uint32_t gguf_get_val_u32 (const struct gguf_context * ctx, int key_id);
GGML_API int32_t gguf_get_val_i32 (const struct gguf_context * ctx, int key_id);
GGML_API float gguf_get_val_f32 (const struct gguf_context * ctx, int key_id);
GGML_API uint64_t gguf_get_val_u64 (const struct gguf_context * ctx, int key_id);
GGML_API int64_t gguf_get_val_i64 (const struct gguf_context * ctx, int key_id);
GGML_API double gguf_get_val_f64 (const struct gguf_context * ctx, int key_id);
GGML_API bool gguf_get_val_bool(const struct gguf_context * ctx, int key_id);
GGML_API const char * gguf_get_val_str (const struct gguf_context * ctx, int key_id);
GGML_API const void * gguf_get_val_data(const struct gguf_context * ctx, int key_id);
GGML_API int gguf_get_arr_n (const struct gguf_context * ctx, int key_id);
GGML_API const void * gguf_get_arr_data(const struct gguf_context * ctx, int key_id);
GGML_API const char * gguf_get_arr_str (const struct gguf_context * ctx, int key_id, int i);
GGML_API int gguf_get_n_tensors (const struct gguf_context * ctx);
GGML_API int gguf_find_tensor (const struct gguf_context * ctx, const char * name);
GGML_API size_t gguf_get_tensor_offset(const struct gguf_context * ctx, int i);
GGML_API char * gguf_get_tensor_name (const struct gguf_context * ctx, int i);
GGML_API enum ggml_type gguf_get_tensor_type (const struct gguf_context * ctx, int i);
// removes key if it exists
GGML_API void gguf_remove_key(struct gguf_context * ctx, const char * key);
// overrides existing values or adds a new one
GGML_API void gguf_set_val_u8 (struct gguf_context * ctx, const char * key, uint8_t val);
GGML_API void gguf_set_val_i8 (struct gguf_context * ctx, const char * key, int8_t val);
GGML_API void gguf_set_val_u16 (struct gguf_context * ctx, const char * key, uint16_t val);
GGML_API void gguf_set_val_i16 (struct gguf_context * ctx, const char * key, int16_t val);
GGML_API void gguf_set_val_u32 (struct gguf_context * ctx, const char * key, uint32_t val);
GGML_API void gguf_set_val_i32 (struct gguf_context * ctx, const char * key, int32_t val);
GGML_API void gguf_set_val_f32 (struct gguf_context * ctx, const char * key, float val);
GGML_API void gguf_set_val_u64 (struct gguf_context * ctx, const char * key, uint64_t val);
GGML_API void gguf_set_val_i64 (struct gguf_context * ctx, const char * key, int64_t val);
GGML_API void gguf_set_val_f64 (struct gguf_context * ctx, const char * key, double val);
GGML_API void gguf_set_val_bool(struct gguf_context * ctx, const char * key, bool val);
GGML_API void gguf_set_val_str (struct gguf_context * ctx, const char * key, const char * val);
GGML_API void gguf_set_arr_data(struct gguf_context * ctx, const char * key, enum gguf_type type, const void * data, int n);
GGML_API void gguf_set_arr_str (struct gguf_context * ctx, const char * key, const char ** data, int n);
// set or add KV pairs from another context
GGML_API void gguf_set_kv(struct gguf_context * ctx, struct gguf_context * src);
// manage tensor info
GGML_API void gguf_add_tensor(struct gguf_context * ctx, const struct ggml_tensor * tensor);
GGML_API void gguf_set_tensor_type(struct gguf_context * ctx, const char * name, enum ggml_type type);
GGML_API void gguf_set_tensor_data(struct gguf_context * ctx, const char * name, const void * data, size_t size);
// writing gguf files can be done in 2 ways:
//
// - write the entire gguf_context to a binary file in a single pass:
//
// gguf_write_to_file(ctx, fname);
//
// - first prepare a file with a placeholder for the meta data, write the tensor data, then write the meta data:
//
// FILE * f = fopen(fname, "wb");
// fseek(f, gguf_get_meta_size(ctx), SEEK_SET);
// fwrite(f, ...);
// void * data = gguf_meta_get_meta_data(ctx);
// fseek(f, 0, SEEK_SET);
// fwrite(f, data, gguf_get_meta_size(ctx));
// free(data);
// fclose(f);
//
// write the entire context to a binary file
GGML_API void gguf_write_to_file(const struct gguf_context * ctx, const char * fname, bool only_meta);
// get the size in bytes of the meta data (header, kv pairs, tensor info) including padding
GGML_API size_t gguf_get_meta_size(const struct gguf_context * ctx);
GGML_API void gguf_get_meta_data(const struct gguf_context * ctx, void * data);
#ifdef __cplusplus
// restrict not standard in C++
#define GGML_RESTRICT
#ifdef __cplusplus
// restrict not standard in C++
# if defined(__GNUC__)
# define GGML_RESTRICT __restrict__
# elif defined(__clang__)
# define GGML_RESTRICT __restrict
# elif defined(_MSC_VER)
# define GGML_RESTRICT __restrict
# else
# define GGML_RESTRICT
# endif
#else
#define GGML_RESTRICT restrict
# define GGML_RESTRICT restrict
#endif
typedef void (*ggml_to_float_t) (const void * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
typedef void (*ggml_from_float_t)(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);

202
ggml/include/gguf.h Normal file
View File

@@ -0,0 +1,202 @@
// This file contains functionality related to "GGUF" files, the binary file format used by ggml.
// GGUF files have the following structure:
//
// 1. File magic "GGUF" (4 bytes).
// 2. File version (uint32_t).
// 3. Number of ggml tensors in file (int64_t).
// 4. Number of key-value-pairs in file (int64_t).
// 5. For each KV pair:
// 1. The key (string).
// 2. The value type (gguf_type).
// 3a. If the value type is GGUF_TYPE_ARRAY:
// 1. The type of the array (gguf_type).
// 2. The number of elements in the array (uint64_t).
// 3. The binary representation of each element in the array.
// 3b. Otherwise:
// 1. The binary representation of the value.
// 6. For each ggml tensor:
// 1. The tensor name (string).
// 2. The number of dimensions of the tensor (uint32_t).
// 3. For each dimension:
// 1. The size of the tensor in the dimension (int64_t).
// 4. The tensor data type (ggml_type).
// 5. The tensor data offset in the tensor data binary blob (uint64_t).
// 7. The tensor data binary blob (optional, aligned).
//
// Strings are serialized as the string length (uint64_t) followed by the C string without the null terminator.
// All enums are stored as int32_t.
// All bool values are stored as int8_t.
// If the special key "general.alignment" (uint32_t) is defined it is used for alignment,
// otherwise GGUF_DEFAULT_ALIGNMENT is used.
//
// Module maintainer: Johannes Gäßler (@JohannesGaessler, johannesg@5d6.de)
#pragma once
#include "ggml.h"
#include <stdbool.h>
#include <stdint.h>
#define GGUF_MAGIC "GGUF"
#define GGUF_VERSION 3
#define GGUF_KEY_GENERAL_ALIGNMENT "general.alignment"
#define GGUF_DEFAULT_ALIGNMENT 32
#ifdef __cplusplus
extern "C" {
#endif
// types that can be stored as GGUF KV data
enum gguf_type {
GGUF_TYPE_UINT8 = 0,
GGUF_TYPE_INT8 = 1,
GGUF_TYPE_UINT16 = 2,
GGUF_TYPE_INT16 = 3,
GGUF_TYPE_UINT32 = 4,
GGUF_TYPE_INT32 = 5,
GGUF_TYPE_FLOAT32 = 6,
GGUF_TYPE_BOOL = 7,
GGUF_TYPE_STRING = 8,
GGUF_TYPE_ARRAY = 9,
GGUF_TYPE_UINT64 = 10,
GGUF_TYPE_INT64 = 11,
GGUF_TYPE_FLOAT64 = 12,
GGUF_TYPE_COUNT, // marks the end of the enum
};
struct gguf_context;
struct gguf_init_params {
bool no_alloc;
// if not NULL, create a ggml_context and allocate the tensor data in it
struct ggml_context ** ctx;
};
GGML_API struct gguf_context * gguf_init_empty(void);
GGML_API struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_params params);
//GGML_API struct gguf_context * gguf_init_from_buffer(..);
GGML_API void gguf_free(struct gguf_context * ctx);
GGML_API const char * gguf_type_name(enum gguf_type type);
GGML_API uint32_t gguf_get_version (const struct gguf_context * ctx);
GGML_API size_t gguf_get_alignment (const struct gguf_context * ctx);
GGML_API size_t gguf_get_data_offset(const struct gguf_context * ctx);
GGML_API int64_t gguf_get_n_kv(const struct gguf_context * ctx);
GGML_API int64_t gguf_find_key(const struct gguf_context * ctx, const char * key); // returns -1 if key is not found
GGML_API const char * gguf_get_key (const struct gguf_context * ctx, int64_t key_id);
GGML_API enum gguf_type gguf_get_kv_type (const struct gguf_context * ctx, int64_t key_id);
GGML_API enum gguf_type gguf_get_arr_type(const struct gguf_context * ctx, int64_t key_id);
// will abort if the wrong type is used for the key
GGML_API uint8_t gguf_get_val_u8 (const struct gguf_context * ctx, int64_t key_id);
GGML_API int8_t gguf_get_val_i8 (const struct gguf_context * ctx, int64_t key_id);
GGML_API uint16_t gguf_get_val_u16 (const struct gguf_context * ctx, int64_t key_id);
GGML_API int16_t gguf_get_val_i16 (const struct gguf_context * ctx, int64_t key_id);
GGML_API uint32_t gguf_get_val_u32 (const struct gguf_context * ctx, int64_t key_id);
GGML_API int32_t gguf_get_val_i32 (const struct gguf_context * ctx, int64_t key_id);
GGML_API float gguf_get_val_f32 (const struct gguf_context * ctx, int64_t key_id);
GGML_API uint64_t gguf_get_val_u64 (const struct gguf_context * ctx, int64_t key_id);
GGML_API int64_t gguf_get_val_i64 (const struct gguf_context * ctx, int64_t key_id);
GGML_API double gguf_get_val_f64 (const struct gguf_context * ctx, int64_t key_id);
GGML_API bool gguf_get_val_bool(const struct gguf_context * ctx, int64_t key_id);
GGML_API const char * gguf_get_val_str (const struct gguf_context * ctx, int64_t key_id);
GGML_API const void * gguf_get_val_data(const struct gguf_context * ctx, int64_t key_id);
GGML_API size_t gguf_get_arr_n (const struct gguf_context * ctx, int64_t key_id);
// get raw pointer to the first element of the array with the given key_id
// for bool arrays, note that they are always stored as int8 on all platforms (usually this makes no difference)
GGML_API const void * gguf_get_arr_data(const struct gguf_context * ctx, int64_t key_id);
// get ith C string from array with given key_id
GGML_API const char * gguf_get_arr_str (const struct gguf_context * ctx, int64_t key_id, size_t i);
GGML_API int64_t gguf_get_n_tensors (const struct gguf_context * ctx);
GGML_API int64_t gguf_find_tensor (const struct gguf_context * ctx, const char * name); // returns -1 if the tensor is not found
GGML_API size_t gguf_get_tensor_offset(const struct gguf_context * ctx, int64_t tensor_id);
GGML_API const char * gguf_get_tensor_name (const struct gguf_context * ctx, int64_t tensor_id);
GGML_API enum ggml_type gguf_get_tensor_type (const struct gguf_context * ctx, int64_t tensor_id);
GGML_API size_t gguf_get_tensor_size (const struct gguf_context * ctx, int64_t tensor_id);
// removes key if it exists, returns id that the key had prior to removal (-1 if it didn't exist)
GGML_API int64_t gguf_remove_key(struct gguf_context * ctx, const char * key);
// overrides an existing KV pair or adds a new one, the new KV pair is always at the back
GGML_API void gguf_set_val_u8 (struct gguf_context * ctx, const char * key, uint8_t val);
GGML_API void gguf_set_val_i8 (struct gguf_context * ctx, const char * key, int8_t val);
GGML_API void gguf_set_val_u16 (struct gguf_context * ctx, const char * key, uint16_t val);
GGML_API void gguf_set_val_i16 (struct gguf_context * ctx, const char * key, int16_t val);
GGML_API void gguf_set_val_u32 (struct gguf_context * ctx, const char * key, uint32_t val);
GGML_API void gguf_set_val_i32 (struct gguf_context * ctx, const char * key, int32_t val);
GGML_API void gguf_set_val_f32 (struct gguf_context * ctx, const char * key, float val);
GGML_API void gguf_set_val_u64 (struct gguf_context * ctx, const char * key, uint64_t val);
GGML_API void gguf_set_val_i64 (struct gguf_context * ctx, const char * key, int64_t val);
GGML_API void gguf_set_val_f64 (struct gguf_context * ctx, const char * key, double val);
GGML_API void gguf_set_val_bool(struct gguf_context * ctx, const char * key, bool val);
GGML_API void gguf_set_val_str (struct gguf_context * ctx, const char * key, const char * val);
// creates a new array with n elements of the given type and copies the corresponding number of bytes from data
GGML_API void gguf_set_arr_data(struct gguf_context * ctx, const char * key, enum gguf_type type, const void * data, size_t n);
// creates a new array with n strings and copies the corresponding strings from data
GGML_API void gguf_set_arr_str (struct gguf_context * ctx, const char * key, const char ** data, size_t n);
// set or add KV pairs from another context
GGML_API void gguf_set_kv(struct gguf_context * ctx, const struct gguf_context * src);
// add tensor to GGUF context, tensor name must be unique
GGML_API void gguf_add_tensor(struct gguf_context * ctx, const struct ggml_tensor * tensor);
// after changing a tensor's type, the offsets of all tensors with higher indices are immediately recalculated
// in such a way that the tensor data remains as one contiguous block (except for padding)
GGML_API void gguf_set_tensor_type(struct gguf_context * ctx, const char * name, enum ggml_type type);
// assumes that at least gguf_get_tensor_size bytes can be read from data
GGML_API void gguf_set_tensor_data(struct gguf_context * ctx, const char * name, const void * data);
// writing gguf files can be done in 3 ways:
//
// - write the entire gguf_context to a binary file in a single pass:
//
// gguf_write_to_file(ctx, fname, /*only_meta =*/ false);
//
// - write only the meta data to a file, then re-open the file and append the tensor data:
//
// gguf_write_to_file(ctx, fname, /*only_meta =*/ true);
// FILE * f = fopen(fname, "ab");
// fwrite(f, ...); // write tensor data
// fclose(f);
//
// - first prepare a file with a placeholder for the meta data, write the tensor data, then write the meta data:
//
// FILE * f = fopen(fname, "wb");
// const size_t size_meta = gguf_get_meta_size(ctx);
// fseek(f, size_meta, SEEK_SET);
// fwrite(f, ...); // write tensor data
// void * data = malloc(size_meta);
// gguf_get_meta_data(ctx, data);
// rewind(f);
// fwrite(data, 1, data, f);
// free(data);
// fclose(f);
//
// write the entire context to a binary file
GGML_API bool gguf_write_to_file(const struct gguf_context * ctx, const char * fname, bool only_meta);
// get the size in bytes of the meta data (header, kv pairs, tensor info) including padding
GGML_API size_t gguf_get_meta_size(const struct gguf_context * ctx);
// writes the meta data to pointer "data"
GGML_API void gguf_get_meta_data(const struct gguf_context * ctx, void * data);
#ifdef __cplusplus
}
#endif