Use main project C files

pull/277/head
Kevin 3 years ago
parent d0060e8910
commit db5f7b7b56

@ -0,0 +1,15 @@
*.iml
.gradle
/local.properties
/.idea/caches
/.idea/libraries
/.idea/modules.xml
/.idea/workspace.xml
/.idea/navEditor.xml
/.idea/assetWizardSettings.xml
.DS_Store
/build
/captures
.externalNativeBuild
.cxx
local.properties

@ -0,0 +1,3 @@
# Default ignored files
/shelf/
/workspace.xml

@ -0,0 +1,6 @@
<?xml version="1.0" encoding="UTF-8"?>
<project version="4">
<component name="CompilerConfiguration">
<bytecodeTargetLevel target="11" />
</component>
</project>

@ -0,0 +1,18 @@
<?xml version="1.0" encoding="UTF-8"?>
<project version="4">
<component name="GradleSettings">
<option name="linkedExternalProjectsSettings">
<GradleProjectSettings>
<option name="testRunner" value="GRADLE" />
<option name="distributionType" value="DEFAULT_WRAPPED" />
<option name="externalProjectPath" value="$PROJECT_DIR$" />
<option name="modules">
<set>
<option value="$PROJECT_DIR$" />
<option value="$PROJECT_DIR$/app" />
</set>
</option>
</GradleProjectSettings>
</option>
</component>
</project>

@ -0,0 +1,10 @@
<?xml version="1.0" encoding="UTF-8"?>
<project version="4">
<component name="ExternalStorageConfigurationManager" enabled="true" />
<component name="ProjectRootManager" version="2" languageLevel="JDK_11" default="true" project-jdk-name="Android Studio default JDK" project-jdk-type="JavaSDK">
<output url="file://$PROJECT_DIR$/build/classes" />
</component>
<component name="ProjectType">
<option name="id" value="Android" />
</component>
</project>

@ -0,0 +1,6 @@
<?xml version="1.0" encoding="UTF-8"?>
<project version="4">
<component name="VcsDirectoryMappings">
<mapping directory="$PROJECT_DIR$/../.." vcs="Git" />
</component>
</project>

@ -1,5 +1,6 @@
LOCAL_PATH := $(call my-dir)
include $(CLEAR_VARS)
PROJECT_DIR := $(LOCAL_PATH)/../../../../../../../
WHISPER_LIB_DIR := libwhisper
LOCAL_LDLIBS := -llog
LOCAL_MODULE := libwhisper
@ -13,10 +14,10 @@ ifneq ($(APP_OPTIM),debug)
LOCAL_LDFLAGS += -flto
endif
LOCAL_CFLAGS += -DSTDC_HEADERS -std=c11 -I $(WHISPER_LIB_DIR)
LOCAL_CFLAGS += -DSTDC_HEADERS -std=c11 -I $(PROJECT_DIR)
LOCAL_CPPFLAGS += -std=c++11
LOCAL_SRC_FILES := $(WHISPER_LIB_DIR)/ggml.c \
$(WHISPER_LIB_DIR)/whisper.cpp \
LOCAL_SRC_FILES := $(PROJECT_DIR)/ggml.c \
$(PROJECT_DIR)/whisper.cpp \
$(LOCAL_PATH)/jni.c
include $(BUILD_SHARED_LIBRARY)

@ -2,7 +2,7 @@
#include <android/log.h>
#include <stdlib.h>
#include <sys/sysinfo.h>
#include "libwhisper/whisper.h"
#include "whisper.h"
#define UNUSED(x) (void)(x)
#define TAG "JNI"

@ -1,737 +0,0 @@
#pragma once
//
// GGML Tensor Library
//
// This documentation is still a work in progress.
// If you wish some specific topics to be covered, feel free to drop a comment:
//
// https://github.com/ggerganov/whisper.cpp/issues/40
//
// ## Overview
//
// This library implements:
//
// - a set of tensor operations
// - automatic differentiation
// - basic optimization algorithms
//
// The aim of this library is to provide a minimalistic approach for various machine learning tasks. This includes,
// but is not limited to, the following:
//
// - linear regression
// - support vector machines
// - neural networks
//
// The library allows the user to define a certain function using the available tensor operations. This function
// definition is represented internally via a computation graph. Each tensor operation in the function definition
// corresponds to a node in the graph. Having the computation graph defined, the user can choose to compute the
// function's value and/or its gradient with respect to the input variables. Optionally, the function can be optimized
// using one of the available optimization algorithms.
//
// For example, here we define the function: f(x) = a*x^2 + b
//
// {
// struct ggml_init_params params = {
// .mem_size = 16*1024*1024,
// .mem_buffer = NULL,
// };
//
// // memory allocation happens here
// struct ggml_context * ctx = ggml_init(params);
//
// struct ggml_tensor * x = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, 1);
//
// ggml_set_param(ctx, x); // x is an input variable
//
// struct ggml_tensor * a = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, 1);
// struct ggml_tensor * b = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, 1);
// struct ggml_tensor * x2 = ggml_mul(ctx, x, x);
// struct ggml_tensor * f = ggml_add(ctx, ggml_mul(ctx, a, x2), b);
//
// ...
// }
//
// Notice that the function definition above does not involve any actual computation. The computation is performed only
// when the user explicitly requests it. For example, to compute the function's value at x = 2.0:
//
// {
// ...
//
// struct ggml_cgraph gf = ggml_build_forward(f);
//
// // set the input variable and parameter values
// ggml_set_f32(x, 2.0f);
// ggml_set_f32(a, 3.0f);
// ggml_set_f32(b, 4.0f);
//
// ggml_graph_compute(ctx0, &gf);
//
// printf("f = %f\n", ggml_get_f32_1d(f, 0));
//
// ...
// }
//
// The actual computation is performed in the ggml_graph_compute() function.
//
// The ggml_new_tensor_...() functions create new tensors. They are allocated in the memory buffer provided to the
// ggml_init() function. You have to be careful not to exceed the memory buffer size. Therefore, you have to know
// in advance how much memory you need for your computation. Alternatively, you can allocate a large enough memory
// and after defining the computation graph, call the ggml_used_mem() function to find out how much memory was
// actually needed.
//
// The ggml_set_param() function marks a tensor as an input variable. This is used by the automatic
// differentiation and optimization algorithms.
//
// The described approach allows to define the function graph once and then compute its forward or backward graphs
// multiple times. All computations will use the same memory buffer allocated in the ggml_init() function. This way
// the user can avoid the memory allocation overhead at runtime.
//
// The library supports multi-dimensional tensors - up to 4 dimensions. The FP16 and FP32 data types are first class
// citizens, but in theory the library can be extended to support FP8 and integer data types.
//
// Each tensor operation produces a new tensor. Initially the library was envisioned to support only the use of unary
// and binary operations. Most of the available operations fall into one of these two categories. With time, it became
// clear that the library needs to support more complex operations. The way to support these operations is not clear
// yet, but a few examples are demonstrated in the following operations:
//
// - ggml_permute()
// - ggml_conv_1d_1s()
// - ggml_conv_1d_2s()
//
// For each tensor operator, the library implements a forward and backward computation function. The forward function
// computes the output tensor value given the input tensor values. The backward function computes the adjoint of the
// input tensors given the adjoint of the output tensor. For a detailed explanation of what this means, take a
// calculus class, or watch the following video:
//
// What is Automatic Differentiation?
// https://www.youtube.com/watch?v=wG_nF1awSSY
//
//
// ## Tensor data (struct ggml_tensor)
//
// The tensors are stored in memory via the ggml_tensor struct. The structure provides information about the size of
// the tensor, the data type, and the memory buffer where the tensor data is stored. Additionally, it contains
// pointers to the "source" tensors - i.e. the tensors that were used to compute the current tensor. For example:
//
// {
// struct ggml_tensor * c = ggml_add(ctx, a, b);
//
// assert(c->src[0] == a);
// assert(c->src[1] == b);
// }
//
// The multi-dimensional tensors are stored in row-major order. The ggml_tensor struct contains fields for the
// number of elements in each dimension ("ne") as well as the number of bytes ("nb", a.k.a. stride). This allows
// to store tensors that are not contiguous in memory, which is useful for operations such as transposition and
// permutation. All tensor operations have to take the stride into account and not assume that the tensor is
// contiguous in memory.
//
// The data of the tensor is accessed via the "data" pointer. For example:
//
// {
// struct ggml_tensor * a = ggml_new_tensor_2d(ctx, GGML_TYPE_F32, 2, 3);
//
// // a[1, 2] = 1.0f;
// *(float *) ((char *) a->data + 2*a->nb[1] + 1*a->nb[0]) = 1.0f;
//
// // a[2, 0] = 2.0f;
// *(float *) ((char *) a->data + 0*a->nb[1] + 2*a->nb[0]) = 2.0f;
//
// ...
// }
//
// Alternatively, there are helper functions, such as ggml_get_f32_1d() and ggml_set_f32_1d() that can be used.
//
// ## The matrix multiplication operator (ggml_mul_mat)
//
// TODO
//
//
// ## Multi-threading
//
// TODO
//
//
// ## Overview of ggml.c
//
// TODO
//
//
// ## SIMD optimizations
//
// TODO
//
//
// ## Debugging ggml
//
// TODO
//
//
#ifdef __cplusplus
extern "C" {
#endif
#include <stdint.h>
#include <stddef.h>
#include <stdbool.h>
#define GGML_MAX_DIMS 4
#define GGML_MAX_NODES 4096
#define GGML_MAX_PARAMS 16
#define GGML_MAX_CONTEXTS 64
#define GGML_MAX_OPT 4
#ifdef __ARM_NEON
// we use the built-in 16-bit float type
typedef __fp16 ggml_fp16_t;
#else
typedef uint16_t ggml_fp16_t;
#endif
// convert FP16 <-> FP32
float ggml_fp16_to_fp32(ggml_fp16_t x);
ggml_fp16_t ggml_fp32_to_fp16(float x);
struct ggml_object;
struct ggml_context;
enum ggml_type {
GGML_TYPE_I8,
GGML_TYPE_I16,
GGML_TYPE_I32,
GGML_TYPE_F16,
GGML_TYPE_F32,
GGML_TYPE_COUNT,
};
// available tensor operations:
enum ggml_op {
GGML_OP_NONE = 0,
GGML_OP_DUP,
GGML_OP_ADD,
GGML_OP_SUB,
GGML_OP_MUL,
GGML_OP_DIV,
GGML_OP_SQR,
GGML_OP_SQRT,
GGML_OP_SUM,
GGML_OP_MEAN,
GGML_OP_REPEAT,
GGML_OP_ABS,
GGML_OP_SGN,
GGML_OP_NEG,
GGML_OP_STEP,
GGML_OP_RELU,
GGML_OP_GELU,
GGML_OP_NORM, // normalize
GGML_OP_MUL_MAT,
GGML_OP_SCALE,
GGML_OP_CPY,
GGML_OP_RESHAPE,
GGML_OP_VIEW,
GGML_OP_PERMUTE,
GGML_OP_TRANSPOSE,
GGML_OP_GET_ROWS,
GGML_OP_DIAG_MASK_INF,
GGML_OP_SOFT_MAX,
GGML_OP_ROPE,
GGML_OP_CONV_1D_1S,
GGML_OP_CONV_1D_2S,
GGML_OP_FLASH_ATTN,
GGML_OP_FLASH_FF,
GGML_OP_COUNT,
};
// n-dimensional tensor
struct ggml_tensor {
enum ggml_type type;
int n_dims;
int ne[GGML_MAX_DIMS]; // number of elements
size_t nb[GGML_MAX_DIMS]; // stride in bytes:
// nb[0] = sizeof(type)
// nb[1] = nb[0] * ne[0] + padding
// nb[i] = nb[i-1] * ne[i-1]
// compute data
enum ggml_op op;
bool is_param;
struct ggml_tensor * grad;
struct ggml_tensor * src0;
struct ggml_tensor * src1;
struct ggml_tensor * opt[GGML_MAX_OPT];
// thread scheduling
int n_tasks;
// performance
int perf_runs;
int64_t perf_cycles;
int64_t perf_time_us;
void * data;
char padding[8];
};
// computation graph
struct ggml_cgraph {
int n_nodes;
int n_leafs;
int n_threads;
size_t work_size;
struct ggml_tensor * work;
struct ggml_tensor * nodes[GGML_MAX_NODES];
struct ggml_tensor * grads[GGML_MAX_NODES];
struct ggml_tensor * leafs[GGML_MAX_NODES];
// performance
int perf_runs;
int64_t perf_cycles;
int64_t perf_time_us;
};
struct ggml_init_params {
// memory pool
size_t mem_size; // bytes
void * mem_buffer; // if NULL, memory will be allocated internally
};
void ggml_time_init(void); // call this once at the beginning of the program
int64_t ggml_time_ms(void);
int64_t ggml_time_us(void);
int64_t ggml_cycles(void);
int64_t ggml_cycles_per_ms(void);
void ggml_print_object (const struct ggml_object * obj);
void ggml_print_objects(const struct ggml_context * ctx);
int ggml_nelements(const struct ggml_tensor * tensor);
size_t ggml_nbytes (const struct ggml_tensor * tensor);
size_t ggml_type_size (enum ggml_type type);
size_t ggml_element_size(const struct ggml_tensor * tensor);
struct ggml_context * ggml_init(struct ggml_init_params params);
void ggml_free(struct ggml_context * ctx);
size_t ggml_used_mem(const struct ggml_context * ctx);
struct ggml_tensor * ggml_new_tensor(
struct ggml_context * ctx,
enum ggml_type type,
int n_dims,
const int *ne);
struct ggml_tensor * ggml_new_tensor_1d(
struct ggml_context * ctx,
enum ggml_type type,
int ne0);
struct ggml_tensor * ggml_new_tensor_2d(
struct ggml_context * ctx,
enum ggml_type type,
int ne0,
int ne1);
struct ggml_tensor * ggml_new_tensor_3d(
struct ggml_context * ctx,
enum ggml_type type,
int ne0,
int ne1,
int ne2);
struct ggml_tensor * ggml_new_tensor_4d(
struct ggml_context * ctx,
enum ggml_type type,
int ne0,
int ne1,
int ne2,
int ne3);
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_dup_tensor (struct ggml_context * ctx, const struct ggml_tensor * src);
struct ggml_tensor * ggml_view_tensor(struct ggml_context * ctx, const struct ggml_tensor * src);
struct ggml_tensor * ggml_set_zero(struct ggml_tensor * tensor);
struct ggml_tensor * ggml_set_i32 (struct ggml_tensor * tensor, int32_t value);
struct ggml_tensor * ggml_set_f32 (struct ggml_tensor * tensor, float value);
int32_t ggml_get_i32_1d(const struct ggml_tensor * tensor, int i);
void ggml_set_i32_1d(const struct ggml_tensor * tensor, int i, int32_t value);
float ggml_get_f32_1d(const struct ggml_tensor * tensor, int i);
void ggml_set_f32_1d(const struct ggml_tensor * tensor, int i, float value);
void * ggml_get_data (const struct ggml_tensor * tensor);
float * ggml_get_data_f32(const struct ggml_tensor * tensor);
//
// operations on tensors with backpropagation
//
struct ggml_tensor * ggml_dup(
struct ggml_context * ctx,
struct ggml_tensor * a);
struct ggml_tensor * ggml_add(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
struct ggml_tensor * ggml_sub(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
struct ggml_tensor * ggml_mul(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
struct ggml_tensor * ggml_div(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
struct ggml_tensor * ggml_sqr(
struct ggml_context * ctx,
struct ggml_tensor * a);
struct ggml_tensor * ggml_sqrt(
struct ggml_context * ctx,
struct ggml_tensor * a);
// return scalar
// TODO: compute sum along rows
struct ggml_tensor * ggml_sum(
struct ggml_context * ctx,
struct ggml_tensor * a);
// mean along rows
struct ggml_tensor * ggml_mean(
struct ggml_context * ctx,
struct ggml_tensor * a);
// if a is the same shape as b, and a is not parameter, return a
// otherwise, return a new tensor: repeat(a) to fit in b
struct ggml_tensor * ggml_repeat(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
struct ggml_tensor * ggml_abs(
struct ggml_context * ctx,
struct ggml_tensor * a);
struct ggml_tensor * ggml_sgn(
struct ggml_context * ctx,
struct ggml_tensor * a);
struct ggml_tensor * ggml_neg(
struct ggml_context * ctx,
struct ggml_tensor * a);
struct ggml_tensor * ggml_step(
struct ggml_context * ctx,
struct ggml_tensor * a);
struct ggml_tensor * ggml_relu(
struct ggml_context * ctx,
struct ggml_tensor * a);
// TODO: double-check this computation is correct
struct ggml_tensor * ggml_gelu(
struct ggml_context * ctx,
struct ggml_tensor * a);
// normalize along rows
// TODO: eps is hardcoded to 1e-5 for now
struct ggml_tensor * ggml_norm(
struct ggml_context * ctx,
struct ggml_tensor * a);
// A: m rows, n columns
// B: p rows, n columns (i.e. we transpose it internally)
// result is m columns, p rows
struct ggml_tensor * ggml_mul_mat(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
//
// operations on tensors without backpropagation
//
// in-place, returns view(a)
struct ggml_tensor * ggml_scale(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
// a -> b, return view(b)
struct ggml_tensor * ggml_cpy(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
// return view(a), b specifies the new shape
// TODO: when we start computing gradient, make a copy instead of view
struct ggml_tensor * ggml_reshape(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
// return view(a)
// TODO: when we start computing gradient, make a copy instead of view
struct ggml_tensor * ggml_reshape_2d(
struct ggml_context * ctx,
struct ggml_tensor * a,
int ne0,
int ne1);
// return view(a)
// TODO: when we start computing gradient, make a copy instead of view
struct ggml_tensor * ggml_reshape_3d(
struct ggml_context * ctx,
struct ggml_tensor * a,
int ne0,
int ne1,
int ne2);
// offset in bytes
struct ggml_tensor * ggml_view_1d(
struct ggml_context * ctx,
struct ggml_tensor * a,
int ne0,
size_t offset);
struct ggml_tensor * ggml_view_2d(
struct ggml_context * ctx,
struct ggml_tensor * a,
int ne0,
int ne1,
size_t nb1, // row stride in bytes
size_t offset);
struct ggml_tensor * ggml_permute(
struct ggml_context * ctx,
struct ggml_tensor * a,
int axis0,
int axis1,
int axis2,
int axis3);
// alias for ggml_permute(ctx, a, 1, 0, 2, 3)
struct ggml_tensor * ggml_transpose(
struct ggml_context * ctx,
struct ggml_tensor * a);
struct ggml_tensor * ggml_get_rows(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
// set elements above the diagonal to -INF
// in-place, returns view(a)
struct ggml_tensor * ggml_diag_mask_inf(
struct ggml_context * ctx,
struct ggml_tensor * a,
int n_past);
// in-place, returns view(a)
struct ggml_tensor * ggml_soft_max(
struct ggml_context * ctx,
struct ggml_tensor * a);
// rotary position embedding
// in-place, returns view(a)
// if mode == 1, skip n_past elements
// TODO: avoid creating a new tensor every time
struct ggml_tensor * ggml_rope(
struct ggml_context * ctx,
struct ggml_tensor * a,
int n_past,
int n_dims,
int mode);
// padding = 1
// TODO: we don't support extra parameters for now
// that's why we are hard-coding the stride, padding, and dilation
// not great ..
struct ggml_tensor * ggml_conv_1d_1s(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
struct ggml_tensor * ggml_conv_1d_2s(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b);
struct ggml_tensor * ggml_flash_attn(
struct ggml_context * ctx,
struct ggml_tensor * q,
struct ggml_tensor * k,
struct ggml_tensor * v,
bool masked);
struct ggml_tensor * ggml_flash_ff(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b0,
struct ggml_tensor * b1,
struct ggml_tensor * c0,
struct ggml_tensor * c1);
//
// automatic differentiation
//
void ggml_set_param(
struct ggml_context * ctx,
struct ggml_tensor * tensor);
void ggml_build_forward_expand(struct ggml_cgraph * cgraph, struct ggml_tensor * tensor);
struct ggml_cgraph ggml_build_forward (struct ggml_tensor * tensor);
struct ggml_cgraph ggml_build_backward(struct ggml_context * ctx, struct ggml_cgraph * gf, bool keep);
void ggml_graph_compute(struct ggml_context * ctx, struct ggml_cgraph * cgraph);
void ggml_graph_reset (struct ggml_cgraph * cgraph);
// print info and performance information for the graph
void ggml_graph_print(const struct ggml_cgraph * cgraph);
// dump the graph into a file using the dot format
void ggml_graph_dump_dot(const struct ggml_cgraph * gb, const struct ggml_cgraph * gf, const char * filename);
//
// optimization
//
// optimization methods
enum ggml_opt_type {
GGML_OPT_ADAM,
GGML_OPT_LBFGS,
};
// linesearch methods
enum ggml_linesearch {
GGML_LINESEARCH_DEFAULT = 1,
GGML_LINESEARCH_BACKTRACKING_ARMIJO = 0,
GGML_LINESEARCH_BACKTRACKING_WOLFE = 1,
GGML_LINESEARCH_BACKTRACKING_STRONG_WOLFE = 2,
};
// optimization return values
enum ggml_opt_result {
GGML_OPT_OK = 0,
GGML_OPT_DID_NOT_CONVERGE,
GGML_OPT_NO_CONTEXT,
GGML_OPT_INVALID_WOLFE,
GGML_OPT_FAIL,
GGML_LINESEARCH_FAIL = -128,
GGML_LINESEARCH_MINIMUM_STEP,
GGML_LINESEARCH_MAXIMUM_STEP,
GGML_LINESEARCH_MAXIMUM_ITERATIONS,
GGML_LINESEARCH_INVALID_PARAMETERS,
};
// optimization parameters
//
// see ggml.c (ggml_opt_default_params) for default values
//
struct ggml_opt_params {
enum ggml_opt_type type;
int n_threads;
// delta-based convergence test
//
// if past == 0 - disabled
// if past > 0:
// stop if |f(x) - f(x_past)| < delta * max(1, |f(x)|)
//
int past;
float delta;
// maximum number of iterations without improvement
//
// if 0 - disabled
// if > 0:
// assume convergence if no cost improvement in this number of iterations
//
int max_no_improvement;
bool print_forward_graph;
bool print_backward_graph;
union {
// ADAM parameters
struct {
int n_iter;
float alpha; // learning rate
float beta1;
float beta2;
float eps; // epsilon for numerical stability
float eps_f; // epsilon for convergence test
float eps_g; // epsilon for convergence test
} adam;
// LBFGS parameters
struct {
int m; // number of corrections to approximate the inv. Hessian
int n_iter;
int max_linesearch;
float eps; // convergence tolerance
float ftol; // line search tolerance
float wolfe;
float min_step;
float max_step;
enum ggml_linesearch linesearch;
} lbfgs;
};
};
struct ggml_opt_params ggml_opt_default_params(enum ggml_opt_type type);
// optimize the function defined by the tensor f
enum ggml_opt_result ggml_opt(
struct ggml_context * ctx,
struct ggml_opt_params params,
struct ggml_tensor * f);
//
// system info
//
int ggml_cpu_has_avx(void);
int ggml_cpu_has_avx2(void);
int ggml_cpu_has_avx512(void);
int ggml_cpu_has_neon(void);
int ggml_cpu_has_f16c(void);
int ggml_cpu_has_fp16_va(void);
int ggml_cpu_has_wasm_simd(void);
int ggml_cpu_has_blas(void);
#ifdef __cplusplus
}
#endif

@ -1,296 +0,0 @@
#ifndef WHISPER_H
#define WHISPER_H
#include <stdint.h>
#include <stdbool.h>
#ifdef WHISPER_SHARED
# ifdef _WIN32
# ifdef WHISPER_BUILD
# define WHISPER_API __declspec(dllexport)
# else
# define WHISPER_API __declspec(dllimport)
# endif
# else
# define WHISPER_API __attribute__ ((visibility ("default")))
# endif
#else
# define WHISPER_API
#endif
#define WHISPER_SAMPLE_RATE 16000
#define WHISPER_N_FFT 400
#define WHISPER_N_MEL 80
#define WHISPER_HOP_LENGTH 160
#define WHISPER_CHUNK_SIZE 30
#ifdef __cplusplus
extern "C" {
#endif
//
// C interface
//
// The following interface is thread-safe as long as the sample whisper_context is not used by multiple threads
// concurrently.
//
// Basic usage:
//
// #include "whisper.h"
//
// ...
//
// struct whisper_context * ctx = whisper_init("/path/to/ggml-base.en.bin");
//
// if (whisper_full(ctx, wparams, pcmf32.data(), pcmf32.size()) != 0) {
// fprintf(stderr, "failed to process audio\n");
// return 7;
// }
//
// const int n_segments = whisper_full_n_segments(ctx);
// for (int i = 0; i < n_segments; ++i) {
// const char * text = whisper_full_get_segment_text(ctx, i);
// printf("%s", text);
// }
//
// whisper_free(ctx);
//
// ...
//
// This is a demonstration of the most straightforward usage of the library.
// "pcmf32" contains the RAW audio data in 32-bit floating point format.
//
// The interface also allows for more fine-grained control over the computation, but it requires a deeper
// understanding of how the model works.
//
struct whisper_context;
typedef int whisper_token;
typedef struct whisper_token_data {
whisper_token id; // token id
whisper_token tid; // forced timestamp token id
float p; // probability of the token
float pt; // probability of the timestamp token
float ptsum; // sum of probabilities of all timestamp tokens
// token-level timestamp data
// do not use if you haven't computed token-level timestamps
int64_t t0; // start time of the token
int64_t t1; // end time of the token
float vlen; // voice length of the token
} whisper_token_data;
// Allocates all memory needed for the model and loads the model from the given file.
// Returns NULL on failure.
WHISPER_API struct whisper_context * whisper_init(const char * path_model);
// Frees all memory allocated by the model.
WHISPER_API void whisper_free(struct whisper_context * ctx);
// Convert RAW PCM audio to log mel spectrogram.
// The resulting spectrogram is stored inside the provided whisper context.
// Returns 0 on success
WHISPER_API int whisper_pcm_to_mel(
struct whisper_context * ctx,
const float * samples,
int n_samples,
int n_threads);
// This can be used to set a custom log mel spectrogram inside the provided whisper context.
// Use this instead of whisper_pcm_to_mel() if you want to provide your own log mel spectrogram.
// n_mel must be 80
// Returns 0 on success
WHISPER_API int whisper_set_mel(
struct whisper_context * ctx,
const float * data,
int n_len,
int n_mel);
// Run the Whisper encoder on the log mel spectrogram stored inside the provided whisper context.
// Make sure to call whisper_pcm_to_mel() or whisper_set_mel() first.
// offset can be used to specify the offset of the first frame in the spectrogram.
// Returns 0 on success
WHISPER_API int whisper_encode(
struct whisper_context * ctx,
int offset,
int n_threads);
// Run the Whisper decoder to obtain the logits and probabilities for the next token.
// Make sure to call whisper_encode() first.
// tokens + n_tokens is the provided context for the decoder.
// n_past is the number of tokens to use from previous decoder calls.
// Returns 0 on success
WHISPER_API int whisper_decode(
struct whisper_context * ctx,
const whisper_token * tokens,
int n_tokens,
int n_past,
int n_threads);
// Token sampling methods.
// These are provided for convenience and can be used after each call to whisper_decode().
// You can also implement your own sampling method using the whisper_get_probs() function.
// whisper_sample_best() returns the token with the highest probability
// whisper_sample_timestamp() returns the most probable timestamp token
WHISPER_API whisper_token_data whisper_sample_best(struct whisper_context * ctx);
WHISPER_API whisper_token_data whisper_sample_timestamp(struct whisper_context * ctx, bool is_initial);
// Return the id of the specified language, returns -1 if not found
WHISPER_API int whisper_lang_id(const char * lang);
WHISPER_API int whisper_n_len (struct whisper_context * ctx); // mel length
WHISPER_API int whisper_n_vocab (struct whisper_context * ctx);
WHISPER_API int whisper_n_text_ctx (struct whisper_context * ctx);
WHISPER_API int whisper_is_multilingual(struct whisper_context * ctx);
// The probabilities for the next token
WHISPER_API float * whisper_get_probs(struct whisper_context * ctx);
// Token Id -> String. Uses the vocabulary in the provided context
WHISPER_API const char * whisper_token_to_str(struct whisper_context * ctx, whisper_token token);
// Special tokens
WHISPER_API whisper_token whisper_token_eot (struct whisper_context * ctx);
WHISPER_API whisper_token whisper_token_sot (struct whisper_context * ctx);
WHISPER_API whisper_token whisper_token_prev(struct whisper_context * ctx);
WHISPER_API whisper_token whisper_token_solm(struct whisper_context * ctx);
WHISPER_API whisper_token whisper_token_not (struct whisper_context * ctx);
WHISPER_API whisper_token whisper_token_beg (struct whisper_context * ctx);
// Task tokens
WHISPER_API whisper_token whisper_token_translate (void);
WHISPER_API whisper_token whisper_token_transcribe(void);
// Performance information
WHISPER_API void whisper_print_timings(struct whisper_context * ctx);
WHISPER_API void whisper_reset_timings(struct whisper_context * ctx);
// Print system information
WHISPER_API const char * whisper_print_system_info(void);
////////////////////////////////////////////////////////////////////////////
// Available sampling strategies
enum whisper_sampling_strategy {
WHISPER_SAMPLING_GREEDY, // Always select the most probable token
WHISPER_SAMPLING_BEAM_SEARCH, // TODO: not implemented yet!
};
// Text segment callback
// Called on every newly generated text segment
// Use the whisper_full_...() functions to obtain the text segments
typedef void (*whisper_new_segment_callback)(struct whisper_context * ctx, int n_new, void * user_data);
// Encoder begin callback
// If not NULL, called before the encoder starts
// If it returns false, the computation is aborted
typedef bool (*whisper_encoder_begin_callback)(struct whisper_context * ctx, void * user_data);
// Parameters for the whisper_full() function
// If you chnage the order or add new parameters, make sure to update the default values in whisper.cpp:
// whisper_full_default_params()
struct whisper_full_params {
enum whisper_sampling_strategy strategy;
int n_threads;
int n_max_text_ctx;
int offset_ms; // start offset in ms
int duration_ms; // audio duration to process in ms
bool translate;
bool no_context;
bool single_segment; // force single segment output (useful for streaming)
bool print_special;
bool print_progress;
bool print_realtime;
bool print_timestamps;
// [EXPERIMENTAL] token-level timestamps
bool token_timestamps; // enable token-level timestamps
float thold_pt; // timestamp token probability threshold (~0.01)
float thold_ptsum; // timestamp token sum probability threshold (~0.01)
int max_len; // max segment length in characters
int max_tokens; // max tokens per segment (0 = no limit)
// [EXPERIMENTAL] speed-up techniques
bool speed_up; // speed-up the audio by 2x using Phase Vocoder
int audio_ctx; // overwrite the audio context size (0 = use default)
// tokens to provide the whisper model as initial prompt
// these are prepended to any existing text context from a previous call
const whisper_token * prompt_tokens;
int prompt_n_tokens;
const char * language;
struct {
int n_past;
} greedy;
struct {
int n_past;
int beam_width;
int n_best;
} beam_search;
whisper_new_segment_callback new_segment_callback;
void * new_segment_callback_user_data;
whisper_encoder_begin_callback encoder_begin_callback;
void * encoder_begin_callback_user_data;
};
WHISPER_API struct whisper_full_params whisper_full_default_params(enum whisper_sampling_strategy strategy);
// Run the entire model: PCM -> log mel spectrogram -> encoder -> decoder -> text
// Uses the specified decoding strategy to obtain the text.
WHISPER_API int whisper_full(
struct whisper_context * ctx,
struct whisper_full_params params,
const float * samples,
int n_samples);
// Split the input audio in chunks and process each chunk separately using whisper_full()
// It seems this approach can offer some speedup in some cases.
// However, the transcription accuracy can be worse at the beginning and end of each chunk.
WHISPER_API int whisper_full_parallel(
struct whisper_context * ctx,
struct whisper_full_params params,
const float * samples,
int n_samples,
int n_processors);
// Number of generated text segments.
// A segment can be a few words, a sentence, or even a paragraph.
WHISPER_API int whisper_full_n_segments(struct whisper_context * ctx);
// Get the start and end time of the specified segment.
WHISPER_API int64_t whisper_full_get_segment_t0(struct whisper_context * ctx, int i_segment);
WHISPER_API int64_t whisper_full_get_segment_t1(struct whisper_context * ctx, int i_segment);
// Get the text of the specified segment.
WHISPER_API const char * whisper_full_get_segment_text(struct whisper_context * ctx, int i_segment);
// Get number of tokens in the specified segment.
WHISPER_API int whisper_full_n_tokens(struct whisper_context * ctx, int i_segment);
// Get the token text of the specified token in the specified segment.
WHISPER_API const char * whisper_full_get_token_text(struct whisper_context * ctx, int i_segment, int i_token);
WHISPER_API whisper_token whisper_full_get_token_id (struct whisper_context * ctx, int i_segment, int i_token);
// Get token data for the specified token in the specified segment.
// This contains probabilities, timestamps, etc.
WHISPER_API whisper_token_data whisper_full_get_token_data(struct whisper_context * ctx, int i_segment, int i_token);
// Get the probability of the specified token in the specified segment.
WHISPER_API float whisper_full_get_token_p(struct whisper_context * ctx, int i_segment, int i_token);
#ifdef __cplusplus
}
#endif
#endif
Loading…
Cancel
Save