mirror of
https://github.com/ollama/ollama.git
synced 2025-04-13 14:19:23 +02:00
update comment
This commit is contained in:
parent
a1c8b0fdb0
commit
62108621d5
@ -77,7 +77,8 @@ func (m *Model) EncodeMultimodal(ctx ml.Context, multimodalData []byte) (any, er
|
||||
// PostTokenize arranges Mistral 3's inputs for the forward pass
|
||||
// In Mistral 3 and Pixtral, the input patches are arranged as follows:
|
||||
// [IMG]...[IMG][IMG_BREAK][IMG]...[IMG][IMG_BREAK][IMG]...[IMG][IMG_END]
|
||||
// Each sequence of [IMG]...[IMG] is a single patch or "row" of vision embeddings
|
||||
// Each sequence of [IMG]...[IMG] is a set of patches of vision embeddings
|
||||
// that can be processed together.
|
||||
func (m *Model) PostTokenize(inputs []input.Input) ([]input.Input, error) {
|
||||
var result []input.Input
|
||||
|
||||
|
Loading…
x
Reference in New Issue
Block a user