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
//! Error types for the inference runtime.
use thiserror::Error;
/// Result type alias for runtime operations.
pub type RuntimeResult<T> = Result<T, RuntimeError>;
/// Errors that can occur during inference.
#[derive(Error, Debug)]
pub enum RuntimeError {
/// No model has been loaded yet.
#[error("no model loaded")]
ModelNotLoaded,
/// Tokenizer is not available because neither `tokenizer-wasm` nor `tokenizer-onig`
/// feature is enabled.
///
/// Enable the `tokenizer-wasm` feature (default, pure Rust) to use
/// the HuggingFace tokenizers library.
#[error("tokenizer not available: rebuild with the `tokenizer-wasm` feature enabled")]
TokenizerNotAvailable,
/// Tokenizer initialization or encoding/decoding failed.
#[error("tokenizer error: {message}")]
TokenizerError {
/// Description of the tokenizer error.
message: String,
},
/// Sampling operation failed.
#[error("sampling error: {message}")]
SamplingError {
/// Description of the sampling error.
message: String,
},
/// KV cache has reached its maximum capacity.
#[error("KV cache full: maximum context length {max_ctx} reached")]
KvCacheFull {
/// Maximum context length supported.
max_ctx: usize,
},
/// Model file could not be loaded.
#[error("model loading error: {message}")]
ModelLoadError {
/// Description of the loading error.
message: String,
},
/// Generation was interrupted or cancelled.
#[error("generation cancelled")]
Cancelled,
/// Error propagated from architecture layer.
#[error("architecture error: {0}")]
Arch(#[from] oxillama_arch::ArchError),
/// Error propagated from GGUF parser.
#[error("GGUF error: {0}")]
Gguf(#[from] oxillama_gguf::GgufError),
/// Error propagated from quantization kernel.
#[error("quantization error: {0}")]
Quant(#[from] oxillama_quant::QuantError),
/// I/O error during model loading.
#[error("I/O error: {0}")]
Io(#[from] std::io::Error),
/// Grammar error from GBNF constrained sampling.
#[error("grammar error: {0}")]
Grammar(#[from] crate::sampling::grammar::GrammarError),
/// Attention computation error.
#[error("attention error: {message}")]
AttentionError {
/// Description of the attention error.
message: String,
},
/// Snapshot format version is incompatible with this runtime.
#[error("snapshot incompatible: {detail}")]
SnapshotIncompatible {
/// Details about the incompatibility.
detail: String,
},
/// Model fingerprint in snapshot does not match the file on disk.
#[error("model fingerprint mismatch: expected={expected}, found={found}, detail={detail}")]
ModelFingerprintMismatch {
/// The fingerprint expected (from snapshot).
expected: String,
/// The fingerprint found (computed from disk).
found: String,
/// Additional detail about the mismatch.
detail: String,
},
/// Offload pager read past end of backing store.
#[error(
"offload: unexpected EOF at offset {offset}, needed {needed} bytes, {available} available"
)]
OffloadEof {
/// Byte offset at which the read was attempted.
offset: u64,
/// Number of bytes requested.
needed: usize,
/// Number of bytes available from `offset` to end.
available: usize,
},
/// A tensor name was not found in the weight offset map.
#[error("tensor not found in weight map: {0}")]
TensorNotFound(String),
/// An internal RwLock or Mutex was poisoned.
#[error("lock poisoned")]
LockPoisoned,
/// Speculative-engine snapshot format is incompatible with this runtime.
#[error("speculative snapshot incompatible: {0}")]
SpecSnapshotIncompatible(String),
/// Cannot pool an empty sequence (seq_len = 0).
#[error("cannot pool empty sequence")]
EmptySequence,
}