mirrors/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-05-11 18:36:41 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Allow models to force a new batch

Browse source 
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:
Jesse Gross 2025-03-10 20:03:29 -07:00 • committed by Michael Yang
parent a8e83a7654
commit 06007c0a18
4 changed files with 10 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

12
server/prompt.go
View file

@ -26,7 +26,6 @@ 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
@ -41,7 +40,7 @@ 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 || isGemma3) && len(msgs[i].Images) > 1 {
if isMllama && len(msgs[i].Images) > 1 {
return "", nil, errTooManyImages
}
@ -158,12 +157,3 @@ 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
}