mirror of
https://github.com/ollama/ollama.git
synced 2025-04-09 04:18:05 +02:00
Revert "Allow models to force a new batch"
This reverts commit c7eae586b899083acebcd9b3847b89ea78c2850c.
This commit is contained in:
parent
06007c0a18
commit
65b0f329d1
@ -15,12 +15,6 @@ type Input struct {
|
||||
// stored in Multimodal, used for caching and comparing
|
||||
// equality.
|
||||
MultimodalHash uint64
|
||||
|
||||
// BatchBreak forces a new batch to be started with this
|
||||
// input. For example, this can be used to align images
|
||||
// with batches. Note that batches may be divided in additional
|
||||
// locations as well.
|
||||
BatchBreak bool
|
||||
}
|
||||
|
||||
// MultimodalIndex is a multimodal element (such as an image)
|
||||
|
@ -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, BatchBreak: true}, // "<start_of_image>""
|
||||
{Token: 108}, // "\n\n"
|
||||
{Token: 255999}, // "<start_of_image>""
|
||||
}
|
||||
result = append(result, imageInputs...)
|
||||
|
||||
|
@ -363,7 +363,7 @@ func (s *Server) processBatch() error {
|
||||
}
|
||||
}
|
||||
|
||||
if j >= s.batchSize || (inp.BatchBreak && len(seq.pendingInputs) != 0) {
|
||||
if j >= s.batchSize {
|
||||
break
|
||||
}
|
||||
|
||||
|
@ -26,6 +26,7 @@ func chatPrompt(ctx context.Context, m *Model, tokenize tokenizeFunc, opts *api.
|
||||
var system []api.Message
|
||||
|
||||
isMllama := checkMllamaModelFamily(m)
|
||||
isGemma3 := checkGemma3ModelFamily(m)
|
||||
|
||||
var imageNumTokens int
|
||||
// TODO: Ideally we would compute this from the projector metadata but some pieces are implementation dependent
|
||||
@ -40,7 +41,7 @@ func chatPrompt(ctx context.Context, m *Model, tokenize tokenizeFunc, opts *api.
|
||||
n := len(msgs) - 1
|
||||
// in reverse, find all messages that fit into context window
|
||||
for i := n; i >= 0; i-- {
|
||||
if isMllama && len(msgs[i].Images) > 1 {
|
||||
if (isMllama || isGemma3) && len(msgs[i].Images) > 1 {
|
||||
return "", nil, errTooManyImages
|
||||
}
|
||||
|
||||
@ -157,3 +158,12 @@ func checkMllamaModelFamily(m *Model) bool {
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
func checkGemma3ModelFamily(m *Model) bool {
|
||||
for _, arch := range m.Config.ModelFamilies {
|
||||
if arch == "gemma3" {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
Loading…
x
Reference in New Issue
Block a user