Enum NormalizeError 

pub enum NormalizeError {
    Backend(String),
    Load {
        path: PathBuf,
        source: LlamaModelLoadError,
    },
    NotCausal {
        path: PathBuf,
        arch: String,
    },
    Context(LlamaContextLoadError),
    Batch(BatchAddError),
    Decode(DecodeError),
    KvCache(KvCacheConversionError),
}

Everything the normalizer can fail with.

Variants

Backend(String)

llama.cpp’s global backend refused to start. It is a process singleton, so the usual cause is another library in the same process having initialized it first.

Load

The weights could not be opened. crate::prepare checks size and hash before anything reaches here, so a failure at this point is usually a path that names no file.

Fields

path: PathBuf

source: LlamaModelLoadError

NotCausal

The file is not a model this stage runs, refused BEFORE the load.

The check has to happen here, because the alternative is a process abort. llama.cpp loads an encoder-decoder gguf without complaint and then asserts inside the first decode on the cross-attention input a causal decode never sets - llm_graph_input_attn_cross::set_input, measured on a t5 build - and a ggml assert aborts rather than returning, so nothing can catch it after the load. The architecture is read out of the file’s own header first, and a file that declares none is refused on the same grounds: an undeclared arch is a promise nobody can make.

Fields

path: PathBuf

arch: String

Context(LlamaContextLoadError)

A context could not be created for the loaded weights.

Batch(BatchAddError)

The batch would not hold the tokens offered to it.

Decode(DecodeError)

llama.cpp rejected a decode step.

KvCache(KvCacheConversionError)

Rejected draft tokens could not be dropped from the KV cache. Continuing would attend over tokens that were never emitted.

Trait Implementations

impl Debug for NormalizeError

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Display for NormalizeError

fn fmt(&self, __formatter: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Error for NormalizeError

fn source(&self) -> Option<&(dyn Error + 'static)>

Returns the lower-level source of this error, if any.

fn description(&self) -> &str

👎Deprecated since 1.42.0:

use the Display impl or to_string()

fn cause(&self) -> Option<&dyn Error>

👎Deprecated since 1.33.0:

replaced by Error::source, which can support downcasting

fn provide<'a>(&'a self, request: &mut Request<'a>)

🔬This is a nightly-only experimental API. (error_generic_member_access)

Provides type-based access to context intended for error reports.

impl From<BatchAddError> for NormalizeError

fn from(source: BatchAddError) -> Self

Converts to this type from the input type.

impl From<DecodeError> for NormalizeError

fn from(source: DecodeError) -> Self

Converts to this type from the input type.

impl From<KvCacheConversionError> for NormalizeError

fn from(source: KvCacheConversionError) -> Self

Converts to this type from the input type.

impl From<LlamaContextLoadError> for NormalizeError

fn from(source: LlamaContextLoadError) -> Self

Converts to this type from the input type.