mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-21 22:33:56 +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

12
model/models/mistral3/model.go
View File

@@ -133,22 +133,22 @@ func (m *Model) EncodeMultimodal(ctx ml.Context, multimodalData []byte) ([]input
// [IMG]...[IMG][IMG_BREAK][IMG]...[IMG][IMG_BREAK][IMG]...[IMG][IMG_END]
// Each sequence of [IMG]...[IMG] is a set of patches of vision embeddings
// that can be processed together.
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 {
result = append(result, inp)
} else {
for i, row := range inp.Multimodal {
// [IMG]
result = append(result, input.Input{Token: 10, Multimodal: []input.Multimodal{{Tensor: row.Tensor}}, MultimodalHash: inp.MultimodalHash, SameBatch: row.Tensor.Dim(1)})
result = append(result, slices.Repeat([]input.Input{{Token: 10}}, row.Tensor.Dim(1)-1)...)
result = append(result, &input.Input{Token: 10, Multimodal: []input.Multimodal{{Tensor: row.Tensor}}, MultimodalHash: inp.MultimodalHash, SameBatch: row.Tensor.Dim(1)})
result = append(result, slices.Repeat([]*input.Input{{Token: 10}}, row.Tensor.Dim(1)-1)...)
if i == len(inp.Multimodal)-1 {
// [IMG_END]
result = append(result, input.Input{Token: 13})
result = append(result, &input.Input{Token: 13})
} else {
// [IMG_BREAK]
result = append(result, input.Input{Token: 12})
result = append(result, &input.Input{Token: 12})
}
}
}