highperfocused/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-07-12 20:42:59 +02:00
Code Issues Packages Projects Releases Wiki Activity

Revert "Allow models to force a new batch"

Browse Source 
This reverts commit c7eae586b899083acebcd9b3847b89ea78c2850c.
This commit is contained in:
jmorganca
2025-03-11 11:53:37 +01:00
committed by Michael Yang
parent 06007c0a18
commit 65b0f329d1
4 changed files with 14 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
server/prompt.go
View File

@ -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
}