mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-23 23:18:26 +00:00
Code Issues Packages Projects Releases Wiki Activity

perf: build graph for next batch async to keep GPU busy (#11863)

Browse Source 
* perf: build graph for next batch in parallel to keep GPU busy

This refactors the main run loop of the ollama runner to perform the main GPU
intensive tasks (Compute+Floats) in a go routine so we can prepare the next
batch in parallel to reduce the amount of time the GPU stalls waiting for the
next batch of work.

* tests: tune integration tests for ollama engine

This tunes the integration tests to focus more on models supported
by the new engine.
This commit is contained in:
Daniel Hiltgen
2025-08-29 14:20:28 -07:00
committed by GitHub
parent ead4a9a1d0
commit 517807cdf2
20 changed files with 591 additions and 235 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
model/models/gemma3/model.go
View File

@@ -112,8 +112,8 @@ func (m *Model) EncodeMultimodal(ctx ml.Context, multimodalData []byte) ([]input
return []input.Multimodal{{Tensor: visionOutputs}}, nil
}
func (m *Model) PostTokenize(inputs []input.Input) ([]input.Input, error) {
var result []input.Input
func (m *Model) PostTokenize(inputs []*input.Input) ([]*input.Input, error) {
var result []*input.Input
for _, inp := range inputs {
if len(inp.Multimodal) == 0 {
@@ -122,17 +122,17 @@ func (m *Model) PostTokenize(inputs []input.Input) ([]input.Input, error) {
inputMultimodal := inp.Multimodal[0].Tensor
result = append(result,
input.Input{Token: 108, SameBatch: inputMultimodal.Dim(1) + 3}, // "\n\n"
input.Input{Token: 255999}, // "<start_of_image>""
input.Input{Multimodal: []input.Multimodal{{Tensor: inputMultimodal}}, MultimodalHash: inp.MultimodalHash}, // image data is on the first placeholder
&input.Input{Token: 108, SameBatch: inputMultimodal.Dim(1) + 3}, // "\n\n"
&input.Input{Token: 255999}, // "<start_of_image>""
&input.Input{Multimodal: []input.Multimodal{{Tensor: inputMultimodal}}, MultimodalHash: inp.MultimodalHash}, // image data is on the first placeholder
)
// add image token placeholders
result = append(result, slices.Repeat([]input.Input{{Token: 0}}, inputMultimodal.Dim(1)-1)...)
result = append(result, slices.Repeat([]*input.Input{{Token: 0}}, inputMultimodal.Dim(1)-1)...)
result = append(result,
input.Input{Token: 256000}, // <end_of_image>
input.Input{Token: 108}, // "\n\n"
&input.Input{Token: 256000}, // <end_of_image>
&input.Input{Token: 108}, // "\n\n"
)
}
}