highperfocused/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-04-07 11:28:17 +02:00
Code Issues Packages Projects Releases Wiki Activity

gemma3: Allow multiple image in a single input

Browse Source 
Previously processing multiple images in a batch would trigger
segfaults so sending images together was disabled as a way to
mitigate this. The trigger was processing one image on the CPU
and one on the GPU.

This can no longer happen:
 - The vision encoder is now on the GPU so both images would be
   processed on the GPU.
 - We require images to be fully contained in a batch and each
   image including its special tokens is over half the batch size.
   As a result, we will never get two images in the same batch.

Fixes #9731
This commit is contained in:
Jesse Gross 2025-03-12 16:59:23 -07:00 committed by Jesse Gross
parent 282bfaaa95
commit 7bf793a600
1 changed files with 1 additions and 11 deletions
Show Stats 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
}