llama-cpp-sys-2 0.1.108

Low Level Bindings to llama.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186

#include "unary-ops.h"

static inline float op_abs(float x) {
    return fabsf(x);
}

static inline float op_sgn(float x) {
    return (x > 0.f) ? 1.f : ((x < 0.f) ? -1.f : 0.f);
}

static inline float op_neg(float x) {
    return -x;
}

static inline float op_step(float x) {
    return (x > 0.f) ? 1.f : 0.f;
}

static inline float op_tanh(float x) {
    return tanhf(x);
}

static inline float op_elu(float x) {
    return (x > 0.f) ? x : expm1f(x);
}

static inline float op_relu(float x) {
    return (x > 0.f) ? x : 0.f;
}

static inline float op_sigmoid(float x) {
    return 1.f / (1.f + expf(-x));
}

static inline float op_hardsigmoid(float x) {
    return fminf(1.0f, fmaxf(0.0f, (x + 3.0f) / 6.0f));
}

static inline float op_exp(float x) {
    return expf(x);
}

static inline float op_hardswish(float x) {
    return x * fminf(1.0f, fmaxf(0.0f, (x + 3.0f) / 6.0f));
}

static inline float op_sqr(float x) {
    return x * x;
}

static inline float op_sqrt(float x) {
    return sqrtf(x);
}

static inline float op_sin(float x) {
    return sinf(x);
}

static inline float op_cos(float x) {
    return cosf(x);
}

static inline float op_log(float x) {
    return logf(x);
}

template <float (*op)(float), typename src0_t, typename dst_t>
static inline void vec_unary_op(int64_t n, dst_t * y, const src0_t * x) {
    constexpr auto src0_to_f32 = type_conversion_table<src0_t>::to_f32;
    constexpr auto f32_to_dst  = type_conversion_table<dst_t >::from_f32;

    for (int i = 0; i < n; i++) {
        y[i] = f32_to_dst(op(src0_to_f32(x[i])));
    }
}

template <float (*op)(float), typename src0_t, typename dst_t>
static void apply_unary_op(const ggml_compute_params * params, ggml_tensor * dst) {
    const ggml_tensor * src0 = dst->src[0];

    GGML_ASSERT(ggml_is_contiguous_1(src0) && ggml_is_contiguous_1(dst) && ggml_are_same_shape(src0, dst));

    GGML_TENSOR_UNARY_OP_LOCALS

    GGML_ASSERT( nb0 == sizeof(dst_t));
    GGML_ASSERT(nb00 == sizeof(src0_t));

    const auto [ir0, ir1] = get_thread_range(params, src0);

    for (int64_t ir = ir0; ir < ir1; ++ir) {
        const int64_t i03 = ir/(ne02*ne01);
        const int64_t i02 = (ir - i03*ne02*ne01)/ne01;
        const int64_t i01 = (ir - i03*ne02*ne01 - i02*ne01);

        dst_t        * dst_ptr  = (dst_t  *)       ((char *)       dst->data  + i03*nb3  + i02*nb2  + i01*nb1 );
        const src0_t * src0_ptr = (const src0_t *) ((const char *) src0->data + i03*nb03 + i02*nb02 + i01*nb01);

        vec_unary_op<op>(ne0, dst_ptr, src0_ptr);
    }
}

// TODO: Use the 'traits' lookup table (for type conversion fns), instead of a mass of 'if' conditions with long templates
template <float (*op)(float)>
static void unary_op(const ggml_compute_params * params, ggml_tensor * dst) {
    const ggml_tensor * src0 = dst->src[0];

    /*  */ if (src0->type == GGML_TYPE_F32  && dst->type == GGML_TYPE_F32) { // all f32
        apply_unary_op<op, float, float>(params, dst);
    } else if (src0->type == GGML_TYPE_F16  && dst->type == GGML_TYPE_F16) { // all f16
        apply_unary_op<op, ggml_fp16_t, ggml_fp16_t>(params, dst);
    } else if (src0->type == GGML_TYPE_BF16 && dst->type == GGML_TYPE_BF16) { // all bf16
        apply_unary_op<op, ggml_bf16_t, ggml_bf16_t>(params, dst);
    } else if (src0->type == GGML_TYPE_BF16 && dst->type == GGML_TYPE_F32) {
        apply_unary_op<op, ggml_bf16_t, float>(params, dst);
    } else if (src0->type == GGML_TYPE_F16  && dst->type == GGML_TYPE_F32) {
        apply_unary_op<op, ggml_fp16_t, float>(params, dst);
    } else {
        fprintf(stderr, "%s: unsupported types: dst: %s, src0: %s\n", __func__,
            ggml_type_name(dst->type), ggml_type_name(src0->type));
        GGML_ABORT("fatal error");
    }
}

void ggml_compute_forward_abs(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_abs>(params, dst);
}

void ggml_compute_forward_sgn(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_sgn>(params, dst);
}

void ggml_compute_forward_neg(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_neg>(params, dst);
}

void ggml_compute_forward_step(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_step>(params, dst);
}

void ggml_compute_forward_tanh(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_tanh>(params, dst);
}

void ggml_compute_forward_elu(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_elu>(params, dst);
}

void ggml_compute_forward_relu(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_relu>(params, dst);
}

void ggml_compute_forward_sigmoid(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_sigmoid>(params, dst);
}

void ggml_compute_forward_hardsigmoid(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_hardsigmoid>(params, dst);
}

void ggml_compute_forward_exp(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_exp>(params, dst);
}

void ggml_compute_forward_hardswish(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_hardswish>(params, dst);
}

void ggml_compute_forward_sqr(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_sqr>(params, dst);
}

void ggml_compute_forward_sqrt(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_sqrt>(params, dst);
}

void ggml_compute_forward_sin(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_sin>(params, dst);
}

void ggml_compute_forward_cos(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_cos>(params, dst);
}

void ggml_compute_forward_log(const ggml_compute_params * params, ggml_tensor * dst) {
    unary_op<op_log>(params, dst);
}