upstream-archive/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-23 09:10:53 -05:00
Code Issues Packages Projects Releases Wiki Activity

launch: use vram bytes for model recommendations (#15885)

Browse Source
This commit is contained in:
Parth Sareen
2026-04-29 18:40:14 -07:00
committed by GitHub
parent bad32c7244
commit b6447caebc
6 changed files with 36 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
server/model_recommendations.go
View File

@@ -17,9 +17,12 @@ import (
"github.com/ollama/ollama/api"
"github.com/ollama/ollama/envconfig"
"github.com/ollama/ollama/format"
)
const modelRecommendationsURL = "https://ollama.com/api/experimental/model-recommendations"
const (
modelRecommendationsURL = "https://ollama.com/api/experimental/model-recommendations"
)
var (
modelRecommendationsRefreshInterval = 4 * time.Hour
@@ -320,7 +323,6 @@ func validateModelRecommendations(recs []api.ModelRecommendation) ([]api.ModelRe
for _, rec := range recs {
rec.Model = strings.TrimSpace(rec.Model)
rec.Description = strings.TrimSpace(rec.Description)
rec.VRAM = strings.TrimSpace(rec.VRAM)
if rec.Model == "" {
return nil, errors.New("recommendation missing model")
@@ -391,11 +393,11 @@ var defaultModelRecommendations = []api.ModelRecommendation{
{
Model: "gemma4",
Description: "Reasoning and code generation locally",
VRAM: "~16GB",
VRAMBytes: 12 * format.GigaByte,
},
{
Model: "qwen3.5",
Description: "Reasoning, coding, and visual understanding locally",
VRAM: "~11GB",
VRAMBytes: 14 * format.GigaByte,
},
}

15
server/model_recommendations_test.go
View File

@@ -19,6 +19,7 @@ import (
"github.com/ollama/ollama/api"
"github.com/ollama/ollama/envconfig"
"github.com/ollama/ollama/format"
)
func TestModelRecommendationsDefaultOrder(t *testing.T) {
@@ -41,11 +42,11 @@ func TestModelRecommendationsCacheRefreshAppliesServerSideChanges(t *testing.T)
first := []api.ModelRecommendation{
{Model: " first-cloud:cloud ", Description: " first ", ContextLength: 2048, MaxOutputTokens: 512},
{Model: " first-local ", Description: " first local ", VRAM: " ~3GB "},
{Model: " first-local ", Description: " first local ", VRAMBytes: 3 * format.GigaByte},
}
second := []api.ModelRecommendation{
{Model: "second-cloud:cloud", Description: "second", ContextLength: 4096, MaxOutputTokens: 1024},
{Model: "second-local", Description: "second local", VRAM: "~6GB"},
{Model: "second-local", Description: "second local", VRAMBytes: 6 * format.GigaByte},
}
calls := 0
@@ -76,7 +77,7 @@ func TestModelRecommendationsCacheRefreshAppliesServerSideChanges(t *testing.T)
}
if got, want := cache.Get(), []api.ModelRecommendation{
{Model: "first-cloud:cloud", Description: "first", ContextLength: 2048, MaxOutputTokens: 512},
{Model: "first-local", Description: "first local", VRAM: "~3GB"},
{Model: "first-local", Description: "first local", VRAMBytes: 3 * format.GigaByte},
}; !slices.Equal(got, want) {
t.Fatalf("after first refresh recommendations = %#v, want %#v", got, want)
}
@@ -160,7 +161,7 @@ func TestModelRecommendationsCacheRefreshErrorCasesPreserveCurrentData(t *testin
setupModelRecommendationsTestEnv(t, "")
cache := newModelRecommendationsCache()
stable := []api.ModelRecommendation{{Model: "stable-local", Description: "stable desc", VRAM: "~2GB"}}
stable := []api.ModelRecommendation{{Model: "stable-local", Description: "stable desc", VRAMBytes: 2 * format.GigaByte}}
cache.set(stable)
cache.client = &http.Client{Transport: tc.transport}
@@ -211,7 +212,7 @@ func TestModelRecommendationsSnapshotPersistAndLoad(t *testing.T) {
want := []api.ModelRecommendation{
{Model: "persist-cloud:cloud", Description: "persisted", ContextLength: 8192, MaxOutputTokens: 2048},
{Model: "persist-local", Description: "persisted local", VRAM: "~5GB"},
{Model: "persist-local", Description: "persisted local", VRAMBytes: 5 * format.GigaByte},
}
writer := newModelRecommendationsCache()
@@ -256,7 +257,7 @@ func TestValidateModelRecommendationsTrimsAndDropsInvalidCloudEntries(t *testing
input := []api.ModelRecommendation{
{Model: " good-cloud:cloud ", Description: " good cloud ", ContextLength: 1024, MaxOutputTokens: 256},
{Model: "bad-cloud:cloud", Description: "missing limits"},
{Model: " good-local ", Description: " good local ", VRAM: " ~2GB "},
{Model: " good-local ", Description: " good local ", VRAMBytes: 2 * format.GigaByte},
}
got, err := validateModelRecommendations(input)
@@ -266,7 +267,7 @@ func TestValidateModelRecommendationsTrimsAndDropsInvalidCloudEntries(t *testing
want := []api.ModelRecommendation{
{Model: "good-cloud:cloud", Description: "good cloud", ContextLength: 1024, MaxOutputTokens: 256},
{Model: "good-local", Description: "good local", VRAM: "~2GB"},
{Model: "good-local", Description: "good local", VRAMBytes: 2 * format.GigaByte},
}
if !slices.Equal(got, want) {
t.Fatalf("validated recommendations = %#v, want %#v", got, want)