diffusion_rs_backend 0.1.0

Blazingly fast inference of diffusion models.
Documentation 
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

use diffusion_rs_common::core::cuda::cudarc::driver::sys::CUstream;
use half::{bf16, f16};

#[allow(dead_code)]
extern "C" {
    pub(crate) fn dequantize_blockwise_f32_int8(
        code: *const f32,
        a: *const u8,
        absmax: *const f32,
        out: *mut f32,
        blocksize: i32,
        n: i32,
        stream: CUstream,
    );
    pub(crate) fn dequantize_blockwise_f32_fp4(
        code: *const f32,
        a: *const u8,
        absmax: *const f32,
        out: *mut f32,
        blocksize: i32,
        n: i32,
        stream: CUstream,
    );
    pub(crate) fn dequantize_blockwise_f32_nf4(
        code: *const f32,
        a: *const u8,
        absmax: *const f32,
        out: *mut f32,
        blocksize: i32,
        n: i32,
        stream: CUstream,
    );

    pub(crate) fn dequantize_blockwise_f16_int8(
        code: *const f32,
        a: *const u8,
        absmax: *const f32,
        out: *mut f16,
        blocksize: i32,
        n: i32,
        stream: CUstream,
    );
    pub(crate) fn dequantize_blockwise_f16_fp4(
        code: *const f32,
        a: *const u8,
        absmax: *const f32,
        out: *mut f16,
        blocksize: i32,
        n: i32,
        stream: CUstream,
    );
    pub(crate) fn dequantize_blockwise_f16_nf4(
        code: *const f32,
        a: *const u8,
        absmax: *const f32,
        out: *mut f16,
        blocksize: i32,
        n: i32,
        stream: CUstream,
    );

    pub(crate) fn dequantize_blockwise_bf16_int8(
        code: *const f32,
        a: *const u8,
        absmax: *const f32,
        out: *mut bf16,
        blocksize: i32,
        n: i32,
        stream: CUstream,
    );
    pub(crate) fn dequantize_blockwise_bf16_fp4(
        code: *const f32,
        a: *const u8,
        absmax: *const f32,
        out: *mut bf16,
        blocksize: i32,
        n: i32,
        stream: CUstream,
    );
    pub(crate) fn dequantize_blockwise_bf16_nf4(
        code: *const f32,
        a: *const u8,
        absmax: *const f32,
        out: *mut bf16,
        blocksize: i32,
        n: i32,
        stream: CUstream,
    );

    pub(crate) fn dequantize_8bit_kernel_f32(
        weight: *const i8,
        scb: *const f32,
        out: *mut f32,
        row: i32,
        col: i32,
        n: i32,
    );
    pub(crate) fn dequantize_8bit_kernel_f16(
        weight: *const i8,
        scb: *const f32,
        out: *mut f16,
        row: i32,
        col: i32,
        n: i32,
    );
    pub(crate) fn dequantize_8bit_kernel_bf16(
        weight: *const i8,
        scb: *const f32,
        out: *mut bf16,
        row: i32,
        col: i32,
        n: i32,
    );
}