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
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
//! utilities for working with the kv cache
use crate::context::LlamaContext;
use std::ffi::c_int;
use std::num::{NonZeroU8, TryFromIntError};
/// Errors that can occur when attempting to prepare values for the kv cache
#[derive(Debug, Eq, PartialEq, thiserror::Error)]
pub enum KvCacheConversionError {
/// Sequence id conversion to i32 failed
#[error("Provided sequence id is too large for a i32")]
SeqIdTooLarge(#[source] TryFromIntError),
/// Position 0 conversion to i32 failed
#[error("Provided start position is too large for a i32")]
P0TooLarge(#[source] TryFromIntError),
/// Position 1 conversion to i32 failed
#[error("Provided end position is too large for a i32")]
P1TooLarge(#[source] TryFromIntError),
/// The operation is not supported by the current model/context configuration.
#[error("operation not supported by this model: {0}")]
UnsupportedOperation(String),
}
impl LlamaContext<'_> {
/// Copy the cache from one sequence to another.
///
/// # Parameters
///
/// * `src` - The sequence id to copy the cache from.
/// * `dest` - The sequence id to copy the cache to.
/// * `size` - The size of the cache to copy.
pub fn copy_cache(&mut self, src: i32, dest: i32, size: i32) {
let mem = unsafe { llama_cpp_bindings_sys::llama_get_memory(self.context.as_ptr()) };
unsafe { llama_cpp_bindings_sys::llama_memory_seq_cp(mem, src, dest, 0, size) }
}
/// Copy the cache from one sequence to another.
///
/// # Returns
/// A `Result` indicating whether the operation was successful.
///
/// # Parameters
/// * `src` - The sequence id to copy the cache from.
/// * `dest` - The sequence id to copy the cache to.
/// * `p0` - The start position of the cache to clear. If `None`, the entire cache is copied up to `p1`.
/// * `p1` - The end position of the cache to clear. If `None`, the entire cache is copied starting from `p0`.
///
/// # Errors
/// If either position exceeds [`i32::MAX`].
pub fn copy_kv_cache_seq(
&mut self,
src: i32,
dest: i32,
p0: Option<u32>,
p1: Option<u32>,
) -> Result<(), KvCacheConversionError> {
let p0 = p0
.map_or(Ok(-1), i32::try_from)
.map_err(KvCacheConversionError::P0TooLarge)?;
let p1 = p1
.map_or(Ok(-1), i32::try_from)
.map_err(KvCacheConversionError::P1TooLarge)?;
let mem = unsafe { llama_cpp_bindings_sys::llama_get_memory(self.context.as_ptr()) };
unsafe { llama_cpp_bindings_sys::llama_memory_seq_cp(mem, src, dest, p0, p1) };
Ok(())
}
/// Clear the kv cache for the given sequence within the specified range `[p0, p1)`
/// Returns `false` only when partial sequence removals fail. Full sequence removals always succeed.
///
/// # Returns
/// A `Result` indicating whether the operation was successful. If the sequence id or
/// either position exceeds the maximum i32 value, no removal is attempted and an `Err` is returned.
///
/// # Parameters
/// * `src` - The sequence id to clear the cache for. If `None`, matches all sequences
/// * `p0` - The start position of the cache to clear. If `None`, the entire cache is cleared up to `p1`.
/// * `p1` - The end position of the cache to clear. If `None`, the entire cache is cleared from `p0`.
///
/// # Errors
/// If the sequence id or either position exceeds [`i32::MAX`].
pub fn clear_kv_cache_seq(
&mut self,
src: Option<u32>,
p0: Option<u32>,
p1: Option<u32>,
) -> Result<bool, KvCacheConversionError> {
let src = src
.map_or(Ok(-1), i32::try_from)
.map_err(KvCacheConversionError::SeqIdTooLarge)?;
let p0 = p0
.map_or(Ok(-1), i32::try_from)
.map_err(KvCacheConversionError::P0TooLarge)?;
let p1 = p1
.map_or(Ok(-1), i32::try_from)
.map_err(KvCacheConversionError::P1TooLarge)?;
let mem = unsafe { llama_cpp_bindings_sys::llama_get_memory(self.context.as_ptr()) };
Ok(unsafe { llama_cpp_bindings_sys::llama_memory_seq_rm(mem, src, p0, p1) })
}
/// Clear the KV cache, including both metadata and the underlying data buffers.
pub fn clear_kv_cache(&mut self) {
let mem = unsafe { llama_cpp_bindings_sys::llama_get_memory(self.context.as_ptr()) };
let clear_data_buffers = true;
unsafe { llama_cpp_bindings_sys::llama_memory_clear(mem, clear_data_buffers) }
}
/// Removes all tokens that do not belong to the specified sequence
///
/// # Parameters
///
/// * `seq_id` - The sequence id to keep
pub fn kv_cache_seq_keep(&mut self, seq_id: i32) {
let mem = unsafe { llama_cpp_bindings_sys::llama_get_memory(self.context.as_ptr()) };
unsafe { llama_cpp_bindings_sys::llama_memory_seq_keep(mem, seq_id) }
}
/// Adds relative position "delta" to all tokens that belong to the specified sequence and have positions in `[p0, p1)`
/// If the KV cache is `RoPEd`, the KV data is updated accordingly:
/// - lazily on next [`LlamaContext::decode`]
/// - explicitly with [`Self::kv_cache_update`]
///
/// # Returns
/// A `Result` indicating whether the operation was successful.
///
/// # Parameters
///
/// * `seq_id` - The sequence id to update
/// * `p0` - The start position of the cache to update. If `None`, the entire cache is updated up to `p1`.
/// * `p1` - The end position of the cache to update. If `None`, the entire cache is updated starting from `p0`.
/// * `delta` - The relative position to add to the tokens
///
/// # Errors
/// If either position exceeds [`i32::MAX`].
pub fn kv_cache_seq_add(
&mut self,
seq_id: i32,
p0: Option<u32>,
p1: Option<u32>,
delta: i32,
) -> Result<(), KvCacheConversionError> {
let p0 = p0
.map_or(Ok(-1), i32::try_from)
.map_err(KvCacheConversionError::P0TooLarge)?;
let p1 = p1
.map_or(Ok(-1), i32::try_from)
.map_err(KvCacheConversionError::P1TooLarge)?;
let status = unsafe {
llama_cpp_bindings_sys::llama_rs_memory_seq_add(
self.context.as_ptr(),
seq_id,
p0,
p1,
delta,
)
};
if crate::status_is_ok(status) {
Ok(())
} else {
Err(KvCacheConversionError::UnsupportedOperation(format!(
"kv_cache_seq_add failed (status {})",
crate::status_to_i32(status)
)))
}
}
/// Integer division of the positions by factor of `d > 1`
/// If the KV cache is `RoPEd`, the KV data is updated accordingly:
/// - lazily on next [`LlamaContext::decode`]
/// - explicitly with [`Self::kv_cache_update`]
///
/// # Returns
/// A `Result` indicating whether the operation was successful.
///
/// # Parameters
///
/// * `seq_id` - The sequence id to update
/// * `p0` - The start position of the cache to update. If `None`, the entire cache is updated up to `p1`.
/// * `p1` - The end position of the cache to update. If `None`, the entire cache is updated starting from `p0`.
/// * `d` - The factor to divide the positions by
///
/// # Errors
/// If either position exceeds [`i32::MAX`].
pub fn kv_cache_seq_div(
&mut self,
seq_id: i32,
p0: Option<u32>,
p1: Option<u32>,
d: NonZeroU8,
) -> Result<(), KvCacheConversionError> {
let p0 = p0
.map_or(Ok(-1), i32::try_from)
.map_err(KvCacheConversionError::P0TooLarge)?;
let p1 = p1
.map_or(Ok(-1), i32::try_from)
.map_err(KvCacheConversionError::P1TooLarge)?;
let d = c_int::from(d.get());
let status = unsafe {
llama_cpp_bindings_sys::llama_rs_memory_seq_div(
self.context.as_ptr(),
seq_id,
p0,
p1,
d,
)
};
if crate::status_is_ok(status) {
Ok(())
} else {
Err(KvCacheConversionError::UnsupportedOperation(format!(
"kv_cache_seq_div failed (status {})",
crate::status_to_i32(status)
)))
}
}
/// Returns the largest position present in the KV cache for the specified sequence
///
/// # Parameters
///
/// * `seq_id` - The sequence id to get the max position for
#[must_use]
pub fn kv_cache_seq_pos_max(&self, seq_id: i32) -> i32 {
unsafe {
llama_cpp_bindings_sys::llama_rs_memory_seq_pos_max(self.context.as_ptr(), seq_id)
}
}
}