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
187
188
189
190
191
192
193
194

//! A safe wrapper around `llama_context_params`.
use llama_cpp_sys_2::{ggml_type, llama_context_params};
use std::fmt::Debug;
use std::num::NonZeroU32;

/// A rusty wrapper around `rope_scaling_type`.
#[repr(i8)]
#[derive(Copy, Clone, Debug, PartialEq, Eq)]
pub enum RopeScalingType {
    /// The scaling type is unspecified
    Unspecified = -1,
    /// No scaling
    None = 0,
    /// Linear scaling
    Linear = 1,
    /// Yarn scaling
    Yarn = 2,
}

/// Create a `RopeScalingType` from a `c_int` - returns `RopeScalingType::ScalingUnspecified` if
/// the value is not recognized.
impl From<i8> for RopeScalingType {
    fn from(value: i8) -> Self {
        match value {
            0 => Self::None,
            1 => Self::Linear,
            2 => Self::Yarn,
            _ => Self::Unspecified,
        }
    }
}

/// Create a `c_int` from a `RopeScalingType`.
impl From<RopeScalingType> for i8 {
    fn from(value: RopeScalingType) -> Self {
        match value {
            RopeScalingType::None => 0,
            RopeScalingType::Linear => 1,
            RopeScalingType::Yarn => 2,
            RopeScalingType::Unspecified => -1,
        }
    }
}

/// A safe wrapper around `llama_context_params`.
#[derive(Debug, PartialEq)]
#[allow(
    missing_docs,
    clippy::struct_excessive_bools,
    clippy::module_name_repetitions
)]
pub struct LlamaContextParams {
    /// The random seed
    pub seed: u32,
    /// the number of tokens in the context - [`None`] if defined by the model.
    pub n_ctx: Option<NonZeroU32>,
    pub n_batch: u32,
    pub n_threads: u32,
    pub n_threads_batch: u32,
    pub rope_scaling_type: RopeScalingType,
    pub rope_freq_base: f32,
    pub rope_freq_scale: f32,
    pub yarn_ext_factor: f32,
    pub yarn_attn_factor: f32,
    pub yarn_beta_fast: f32,
    pub yarn_beta_slow: f32,
    pub yarn_orig_ctx: u32,
    pub type_k: ggml_type,
    pub type_v: ggml_type,
    pub mul_mat_q: bool,
    pub logits_all: bool,
    pub embedding: bool,
    pub offload_kqv: bool,
    pub cb_eval: llama_cpp_sys_2::ggml_backend_sched_eval_callback,
    pub cb_eval_user_data: *mut std::ffi::c_void,
}

/// Default parameters for `LlamaContext`. (as defined in llama.cpp by `llama_context_default_params`)
/// ```
/// # use std::num::NonZeroU32;
/// use llama_cpp_2::context::params::{LlamaContextParams, RopeScalingType};
/// let params = LlamaContextParams::default();
/// assert_eq!(params.n_ctx, NonZeroU32::new(512), "n_ctx should be 512");
/// assert_eq!(params.rope_scaling_type, RopeScalingType::Unspecified);
/// ```
impl Default for LlamaContextParams {
    fn default() -> Self {
        Self::from(unsafe { llama_cpp_sys_2::llama_context_default_params() })
    }
}

impl From<llama_context_params> for LlamaContextParams {
    fn from(
        llama_context_params {
            seed,
            n_ctx,
            n_batch,
            n_threads,
            n_threads_batch,
            rope_freq_base,
            rope_freq_scale,
            cb_eval,
            cb_eval_user_data,
            type_k,
            type_v,
            mul_mat_q,
            logits_all,
            embedding,
            rope_scaling_type,
            yarn_ext_factor,
            yarn_attn_factor,
            yarn_beta_fast,
            yarn_beta_slow,
            yarn_orig_ctx,
            offload_kqv,
        }: llama_context_params,
    ) -> Self {
        Self {
            seed,
            n_ctx: NonZeroU32::new(n_ctx),
            n_batch,
            n_threads,
            n_threads_batch,
            rope_freq_base,
            rope_freq_scale,
            type_k,
            type_v,
            mul_mat_q,
            logits_all,
            embedding,
            rope_scaling_type: RopeScalingType::from(rope_scaling_type),
            yarn_ext_factor,
            yarn_attn_factor,
            yarn_beta_fast,
            yarn_beta_slow,
            yarn_orig_ctx,
            offload_kqv,
            cb_eval,
            cb_eval_user_data,
        }
    }
}

impl From<LlamaContextParams> for llama_context_params {
    fn from(
        LlamaContextParams {
            seed,
            n_ctx,
            n_batch,
            n_threads,
            n_threads_batch,
            rope_freq_base,
            rope_freq_scale,
            type_k,
            type_v,
            mul_mat_q,
            logits_all,
            embedding,
            rope_scaling_type,
            yarn_ext_factor,
            yarn_attn_factor,
            yarn_beta_fast,
            yarn_beta_slow,
            yarn_orig_ctx,
            offload_kqv,
            cb_eval,
            cb_eval_user_data,
        }: LlamaContextParams,
    ) -> Self {
        llama_context_params {
            seed,
            n_ctx: n_ctx.map_or(0, NonZeroU32::get),
            n_batch,
            n_threads,
            n_threads_batch,
            rope_freq_base,
            rope_freq_scale,
            type_k,
            type_v,
            mul_mat_q,
            logits_all,
            embedding,
            rope_scaling_type: i8::from(rope_scaling_type),
            yarn_ext_factor,
            yarn_attn_factor,
            yarn_beta_fast,
            yarn_beta_slow,
            yarn_orig_ctx,
            offload_kqv,
            cb_eval,
            cb_eval_user_data,
        }
    }
}