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

tests: add single threaded history test (#12295)

Browse Source 
* tests: add single threaded history test

Also tidies up some existing tests to handle more model output variation

* test: add support for testing specific architectures
This commit is contained in:
Daniel Hiltgen
2025-09-22 11:23:14 -07:00
committed by GitHub
parent af060eb250
commit c23e6f4cae
9 changed files with 173 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
integration/model_perf_test.go
View File

@@ -40,6 +40,18 @@ var (
// cat int.log | grep MODEL_PERF_HEADER | head -1| cut -f2- -d: > perf.csv
// cat int.log | grep MODEL_PERF_DATA | cut -f2- -d: >> perf.csv
func TestModelsPerf(t *testing.T) {
if s := os.Getenv("OLLAMA_NEW_ENGINE"); s != "" {
doModelPerfTest(t, ollamaEngineChatModels)
} else {
doModelPerfTest(t, append(ollamaEngineChatModels, llamaRunnerChatModels...))
}
}
func TestLibraryModelsPerf(t *testing.T) {
doModelPerfTest(t, libraryChatModels)
}
func doModelPerfTest(t *testing.T, chatModels []string) {
softTimeout, hardTimeout := getTimeouts(t)
slog.Info("Setting timeouts", "soft", softTimeout, "hard", hardTimeout)
ctx, cancel := context.WithTimeout(context.Background(), hardTimeout)
@@ -65,14 +77,12 @@ func TestModelsPerf(t *testing.T) {
}
longPrompt := "summarize the following: " + string(data)
var chatModels []string
if s := os.Getenv("OLLAMA_NEW_ENGINE"); s != "" {
chatModels = ollamaEngineChatModels
} else {
chatModels = append(ollamaEngineChatModels, llamaRunnerChatModels...)
}
targetArch := os.Getenv("OLLAMA_TEST_ARCHITECTURE")
for _, model := range chatModels {
if !strings.Contains(model, ":") {
model = model + ":latest"
}
t.Run(model, func(t *testing.T) {
if time.Now().Sub(started) > softTimeout {
t.Skip("skipping remaining tests to avoid excessive runtime")
@@ -88,6 +98,9 @@ func TestModelsPerf(t *testing.T) {
}
arch := resp.ModelInfo["general.architecture"].(string)
maxContext = int(resp.ModelInfo[fmt.Sprintf("%s.context_length", arch)].(float64))
if targetArch != "" && arch != targetArch {
t.Skip(fmt.Sprintf("Skipping %s architecture %s != %s", model, arch, targetArch))
}
if maxVram > 0 {
resp, err := client.List(ctx)
@@ -151,8 +164,8 @@ func TestModelsPerf(t *testing.T) {
prompt string
anyResp []string
}{
{"why is the sky blue?", []string{"rayleigh", "scattering", "atmosphere", "nitrogen", "oxygen"}},
{maxPrompt, []string{"shakespeare", "oppression", "sorrows", "gutenberg", "child", "license", "sonnet", "melancholy"}},
{blueSkyPrompt, blueSkyExpected},
{maxPrompt, []string{"shakespeare", "oppression", "sorrows", "gutenberg", "child", "license", "sonnet", "melancholy", "love", "sorrow", "beauty"}},
}
var gpuPercent int
for _, tc := range testCases {
@@ -241,11 +254,12 @@ func TestModelsPerf(t *testing.T) {
}
}
}
// Round the logged prompt count for comparisons across versions/configurations which can vary slightly
fmt.Fprintf(os.Stderr, "MODEL_PERF_HEADER:%s,%s,%s,%s,%s,%s,%s\n",
"MODEL",
"CONTEXT",
"GPU PERCENT",
"PROMPT COUNT",
"APPROX PROMPT COUNT",
"LOAD TIME",
"PROMPT EVAL TPS",
"EVAL TPS",
@@ -254,7 +268,7 @@ func TestModelsPerf(t *testing.T) {
model,
numCtx,
gpuPercent,
resp.PromptEvalCount,
(resp.PromptEvalCount/10)*10,
float64(resp.LoadDuration)/1000000000.0,
float64(resp.PromptEvalCount)/(float64(resp.PromptEvalDuration)/1000000000.0),
float64(resp.EvalCount)/(float64(resp.EvalDuration)/1000000000.0),