mirror of
https://github.com/dogkeeper886/ollama37.git
synced 2025-12-10 15:57:04 +00:00
Allow models to force a new batch
This is useful for a few things: - Work around bugs, such as having 2 images in one batch - Keep the image in a single batch for fully connected attention - Improve performance by not evaluating embeddings multiple times
This commit is contained in:
committed by
Michael Yang
parent
a8e83a7654
commit
06007c0a18
@@ -112,8 +112,8 @@ func (m *Model) PostTokenize(ctx ml.Context, inputs []input.Input) ([]input.Inpu
|
||||
result = append(result, inp)
|
||||
} else {
|
||||
imageInputs := []input.Input{
|
||||
{Token: 108}, // "\n\n"
|
||||
{Token: 255999}, // "<start_of_image>""
|
||||
{Token: 108}, // "\n\n"
|
||||
{Token: 255999, BatchBreak: true}, // "<start_of_image>""
|
||||
}
|
||||
result = append(result, imageInputs...)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user