sipp-sys 0.1.0

Native llama.cpp FFI layer for Sipp
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

#[cxx::bridge(namespace = "sipp::sys")]
pub mod ffi {
    unsafe extern "C++" {
        include!("sipp_cxx.h");

        type NativeRuntime;
        type NativeBatch;
        type CommonSampler;

        fn backend_init();
        fn backend_load_all();
        fn set_llama_log_quiet(quiet: bool);
        fn backend_observability_json(include_details: bool) -> String;
        fn mtmd_default_marker() -> String;

        fn load_native_runtime(
            model_path: &str,
            args_json: &str,
        ) -> Result<UniquePtr<NativeRuntime>>;
        fn n_ctx(self: &NativeRuntime) -> i32;
        fn n_batch(self: &NativeRuntime) -> i32;
        fn n_ubatch(self: &NativeRuntime) -> i32;
        fn n_seq_max(self: &NativeRuntime) -> i32;
        fn n_threads(self: &NativeRuntime) -> i32;
        fn n_threads_batch(self: &NativeRuntime) -> i32;
        fn n_embd_out(self: &NativeRuntime) -> i32;
        fn n_cls_out(self: &NativeRuntime) -> i32;
        fn pooling_type(self: &NativeRuntime) -> i32;
        fn has_encoder(self: &NativeRuntime) -> bool;
        fn has_decoder(self: &NativeRuntime) -> bool;
        fn has_chat_template(self: &NativeRuntime) -> bool;
        fn is_recurrent(self: &NativeRuntime) -> bool;
        fn is_hybrid(self: &NativeRuntime) -> bool;
        fn kv_unified(self: &NativeRuntime) -> bool;
        fn flash_attention(self: &NativeRuntime) -> String;
        fn cache_type_k(self: &NativeRuntime) -> String;
        fn cache_type_v(self: &NativeRuntime) -> String;
        fn chat_template_source(self: &NativeRuntime) -> String;
        fn bos_token(self: &NativeRuntime) -> i32;
        fn eos_token(self: &NativeRuntime) -> i32;
        fn decoder_start_token(self: &NativeRuntime) -> i32;
        fn is_eog(self: &NativeRuntime, token: i32) -> bool;
        fn mtmd_ready(self: &NativeRuntime) -> bool;
        fn tokenize(
            self: &NativeRuntime,
            text: &str,
            add_special: bool,
            parse_special: bool,
        ) -> Result<Vec<i32>>;
        fn token_to_piece(self: &NativeRuntime, token: i32, special: bool) -> Result<String>;
        fn token_to_piece_bytes_into(
            self: &NativeRuntime,
            token: i32,
            special: bool,
            out: &mut Vec<u8>,
        ) -> Result<()>;
        fn apply_chat_template_json(
            self: &NativeRuntime,
            messages_json: &str,
            add_assistant: bool,
        ) -> Result<String>;
        fn decode(self: Pin<&mut NativeRuntime>, batch: &NativeBatch) -> Result<i32>;
        fn encode(self: Pin<&mut NativeRuntime>, batch: &NativeBatch) -> Result<i32>;
        fn synchronize(self: Pin<&mut NativeRuntime>) -> bool;
        fn clear_sequence(self: Pin<&mut NativeRuntime>, seq_id: i32, p0: i32, p1: i32) -> bool;
        fn add_sequence_delta(
            self: Pin<&mut NativeRuntime>,
            seq_id: i32,
            p0: i32,
            p1: i32,
            delta: i32,
        );
        fn embeddings_seq(self: &NativeRuntime, seq_id: i32) -> Result<Vec<f32>>;
        fn state_seq(self: &NativeRuntime, seq_id: i32) -> Result<Vec<u8>>;
        fn set_state_seq(self: Pin<&mut NativeRuntime>, seq_id: i32, data: &[u8]) -> bool;
        fn init_mtmd(
            self: Pin<&mut NativeRuntime>,
            projector_path: &str,
            use_gpu: bool,
            n_threads: i32,
        ) -> bool;
        fn mtmd_support_vision(self: &NativeRuntime) -> bool;
        fn mtmd_eval_images(
            self: Pin<&mut NativeRuntime>,
            prompt: &str,
            image_bytes: &[u8],
            image_sizes: &[i32],
            add_special: bool,
            parse_special: bool,
            n_past: i32,
            seq_id: i32,
            n_batch: i32,
            logits_last: bool,
        ) -> Result<i32>;

        fn make_native_batch() -> UniquePtr<NativeBatch>;
        fn ensure_capacity(
            self: Pin<&mut NativeBatch>,
            max_tokens: i32,
            max_sequences: i32,
        ) -> Result<()>;
        fn reset(self: Pin<&mut NativeBatch>);
        fn add_token(
            self: Pin<&mut NativeBatch>,
            token: i32,
            pos: i32,
            seq_id: i32,
            logits: bool,
        ) -> bool;
        fn n_tokens(self: &NativeBatch) -> i32;
        fn token(self: &NativeBatch, index: i32) -> i32;
        fn pos(self: &NativeBatch, index: i32) -> i32;
        fn seq_id(self: &NativeBatch, index: i32) -> i32;
        fn logits(self: &NativeBatch, index: i32) -> bool;
        fn clear_logits(self: Pin<&mut NativeBatch>);
        fn set_last_logits(self: Pin<&mut NativeBatch>);

        fn create_sampler(
            runtime: &NativeRuntime,
            sampling_json: &str,
            grammar: &str,
            json_schema: &str,
        ) -> Result<UniquePtr<CommonSampler>>;
        fn sampler_accept(self: Pin<&mut CommonSampler>, token: i32, accept_grammar: bool) -> bool;
        fn sampler_reset(self: Pin<&mut CommonSampler>);
        fn sampler_backend_sampling(self: &CommonSampler) -> bool;
        fn sampler_sample(
            sampler: Pin<&mut CommonSampler>,
            runtime: Pin<&mut NativeRuntime>,
            idx: i32,
        ) -> i32;
        fn sampler_attach(
            sampler: Pin<&mut CommonSampler>,
            runtime: Pin<&mut NativeRuntime>,
            seq_id: i32,
        ) -> bool;
        fn sampler_detach(runtime: Pin<&mut NativeRuntime>, seq_id: i32) -> bool;
    }
}

pub use ffi::*;