Struct MtmdInputChunk 

pub struct MtmdInputChunk {
    pub chunk: NonNull<mtmd_input_chunk>,
    pub owned: bool,
}

Safe wrapper around mtmd_input_chunk.

Represents a single chunk of input data, which can be either text tokens, image tokens, or audio tokens. The chunk type determines what kind of data and operations are available.

Fields

chunk: NonNull<mtmd_input_chunk>

Raw pointer to the underlying mtmd_input_chunk.

owned: bool

Implementations

impl MtmdInputChunk

pub fn chunk_type(&self) -> Result<MtmdInputChunkType, MtmdInputChunkTypeError>

Get the type of this chunk

Errors

Returns an error if the chunk type is unknown.

pub fn text_tokens(&self) -> Option<&[LlamaToken]>

Get text tokens from this chunk.

Only valid for text chunks. Returns None for image or audio chunks.

pub fn n_tokens(&self) -> usize

Get the number of tokens in this chunk

pub fn n_positions(&self) -> i32

Get the number of positions in this chunk.

pub fn id(&self) -> Option<String>

Get chunk ID if available.

Returns None for text chunks, may return an ID for image/audio chunks.

pub fn copy(&self) -> Result<Self, MtmdInputChunkError>

Create a copy of this chunk that you own.

Errors

Returns MtmdInputChunkError::NullResult if copying fails.

pub fn eval_single( &self, mtmd_ctx: &MtmdContext, llama_ctx: &LlamaContext<'_>, start_position: llama_pos, seq_id: llama_seq_id, n_batch: i32, logits_last: bool, ) -> Result<llama_pos, MtmdEvalError>

Evaluate this single chunk through the multimodal helper.

Mirrors MtmdInputChunks::eval_chunks but for one chunk at a time, so callers can interleave per-chunk decode with per-chunk bookkeeping (token counting, marker state-machine replay) inside one loop instead of running the helper-level all-chunks eval and a separate ingest pass.

Image chunks are decoded as one llama_decode call inside the helper, so their token count must fit in n_batch. When it would not, the binding refuses the call up front because the C-side GGML_ASSERT(n_tokens_all <= cparams.n_batch) would otherwise abort the process.

Errors

Returns MtmdEvalError::ImageChunkExceedsBatchSize when this is an image chunk whose token count exceeds n_batch. Returns MtmdEvalError::EvalFailure if the underlying encode or decode step fails.

Trait Implementations

impl Debug for MtmdInputChunk

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

Formats the value using the given formatter.

impl Drop for MtmdInputChunk

fn drop(&mut self)

Executes the destructor for this type.

fn pin_drop(self: Pin<&mut Self>)

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

Execute the destructor for this type, but different to Drop::drop, it requires self to be pinned.