tests: add single threaded history test (#12295)

* 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

integration/README.md

@@ -12,3 +12,6 @@ The integration tests have 2 modes of operating.
 
 > [!IMPORTANT]
 > Before running the tests locally without the "test existing" setting, compile ollama from the top of the source tree  `go build .` in addition to GPU support with cmake if applicable on your platform.  The integration tests expect to find an ollama binary at the top of the tree.
+
+
+Many tests use a default small model suitable to run on many systems.  You can override this default model by setting `OLLAMA_TEST_DEFAULT_MODEL`

integration/api_test.go

@@ -22,13 +22,12 @@ func TestAPIGenerate(t *testing.T) {
 	// Set up the test data
 	req := api.GenerateRequest{
 		Model:  smol,
-		Prompt: "why is the sky blue? be brief",
+		Prompt: blueSkyPrompt,
 		Options: map[string]interface{}{
 			"temperature": 0,
 			"seed":        123,
 		},
 	}
-	anyResp := []string{"rayleigh", "scattering"}
 
 	client, _, cleanup := InitServerConnection(ctx, t)
 	defer cleanup()
@@ -120,14 +119,14 @@ func TestAPIGenerate(t *testing.T) {
 				// Verify the response contains the expected data
 				response := buf.String()
 				atLeastOne := false
-				for _, resp := range anyResp {
+				for _, resp := range blueSkyExpected {
 					if strings.Contains(strings.ToLower(response), resp) {
 						atLeastOne = true
 						break
 					}
 				}
 				if !atLeastOne {
-					t.Errorf("none of %v found in %s", anyResp, response)
+					t.Errorf("none of %v found in %s", blueSkyExpected, response)
 				}
 			case <-ctx.Done():
 				t.Error("outer test context done while waiting for generate")
@@ -181,7 +180,7 @@ func TestAPIChat(t *testing.T) {
 		Messages: []api.Message{
 			{
 				Role:    "user",
-				Content: "why is the sky blue?  be brief",
+				Content: blueSkyPrompt,
 			},
 		},
 		Options: map[string]interface{}{
@@ -189,7 +188,6 @@ func TestAPIChat(t *testing.T) {
 			"seed":        123,
 		},
 	}
-	anyResp := []string{"rayleigh", "scattering"}
 
 	client, _, cleanup := InitServerConnection(ctx, t)
 	defer cleanup()
@@ -279,14 +277,14 @@ func TestAPIChat(t *testing.T) {
 				// Verify the response contains the expected data
 				response := buf.String()
 				atLeastOne := false
-				for _, resp := range anyResp {
+				for _, resp := range blueSkyExpected {
 					if strings.Contains(strings.ToLower(response), resp) {
 						atLeastOne = true
 						break
 					}
 				}
 				if !atLeastOne {
-					t.Errorf("none of %v found in %s", anyResp, response)
+					t.Errorf("none of %v found in %s", blueSkyExpected, response)
 				}
 			case <-ctx.Done():
 				t.Error("outer test context done while waiting for chat")

integration/basic_test.go

@@ -19,14 +19,14 @@ func TestBlueSky(t *testing.T) {
 	// Set up the test data
 	req := api.GenerateRequest{
 		Model:  smol,
-		Prompt: "why is the sky blue?",
+		Prompt: blueSkyPrompt,
 		Stream: &stream,
 		Options: map[string]any{
 			"temperature": 0,
 			"seed":        123,
 		},
 	}
-	GenerateTestHelper(ctx, t, req, []string{"rayleigh", "scattering"})
+	GenerateTestHelper(ctx, t, req, blueSkyExpected)
 }
 
 func TestUnicode(t *testing.T) {
@@ -110,12 +110,12 @@ func TestUnicodeModelDir(t *testing.T) {
 
 	req := api.GenerateRequest{
 		Model:  smol,
-		Prompt: "why is the sky blue?",
+		Prompt: blueSkyPrompt,
 		Stream: &stream,
 		Options: map[string]any{
 			"temperature": 0,
 			"seed":        123,
 		},
 	}
-	GenerateTestHelper(ctx, t, req, []string{"rayleigh", "scattering"})
+	GenerateTestHelper(ctx, t, req, blueSkyExpected)
 }

integration/context_test.go

@@ -63,11 +63,11 @@ func TestContextExhaustion(t *testing.T) {
 	if err := PullIfMissing(ctx, client, req.Model); err != nil {
 		t.Fatalf("PullIfMissing failed: %v", err)
 	}
-	DoGenerate(ctx, t, client, req, []string{"once", "upon", "lived", "sunny", "cloudy", "clear", "water"}, 120*time.Second, 10*time.Second)
+	DoGenerate(ctx, t, client, req, []string{"once", "upon", "lived", "sunny", "cloudy", "clear", "water", "time", "travel", "world"}, 120*time.Second, 10*time.Second)
 }
 
 // Send multiple generate requests with prior context and ensure the response is coherant and expected
-func TestGenerateWithHistory(t *testing.T) {
+func TestParallelGenerateWithHistory(t *testing.T) {
 	modelOverride := ollamaEngineChatModels[0] // Most recent ollama engine model
 	req, resp := GenerateRequests()
 	numParallel := 2
@@ -113,8 +113,48 @@ func TestGenerateWithHistory(t *testing.T) {
 	wg.Wait()
 }
 
+// Send generate requests with prior context and ensure the response is coherant and expected
+func TestGenerateWithHistory(t *testing.T) {
+	req := api.GenerateRequest{
+		Model:     smol,
+		Prompt:    rainbowPrompt,
+		Stream:    &stream,
+		KeepAlive: &api.Duration{Duration: 10 * time.Second},
+		Options: map[string]any{
+			"num_ctx": 16384,
+		},
+	}
+
+	softTimeout, hardTimeout := getTimeouts(t)
+	ctx, cancel := context.WithTimeout(context.Background(), hardTimeout)
+	defer cancel()
+	client, _, cleanup := InitServerConnection(ctx, t)
+	defer cleanup()
+
+	// Get the server running (if applicable) warm the model up with a single initial request
+	slog.Info("loading", "model", req.Model)
+	err := client.Generate(ctx,
+		&api.GenerateRequest{Model: req.Model, KeepAlive: &api.Duration{Duration: 10 * time.Second}, Options: req.Options},
+		func(response api.GenerateResponse) error { return nil },
+	)
+	if err != nil {
+		t.Fatalf("failed to load model %s: %s", req.Model, err)
+	}
+
+	req.Context = DoGenerate(ctx, t, client, req, rainbowExpected, 30*time.Second, 20*time.Second)
+
+	for i := 0; i < len(rainbowFollowups); i++ {
+		req.Prompt = rainbowFollowups[i]
+		if time.Now().Sub(started) > softTimeout {
+			slog.Info("exceeded soft timeout, winding down test")
+			return
+		}
+		req.Context = DoGenerate(ctx, t, client, req, rainbowExpected, 30*time.Second, 20*time.Second)
+	}
+}
+
 // Send multiple chat requests with prior context and ensure the response is coherant and expected
-func TestChatWithHistory(t *testing.T) {
+func TestParallelChatWithHistory(t *testing.T) {
 	modelOverride := ollamaEngineChatModels[0] // Most recent ollama engine model
 	req, resp := ChatRequests()
 	numParallel := 2
@@ -164,3 +204,55 @@ func TestChatWithHistory(t *testing.T) {
 	}
 	wg.Wait()
 }
+
+// Send generate requests with prior context and ensure the response is coherant and expected
+func TestChatWithHistory(t *testing.T) {
+	req := api.ChatRequest{
+		Model:     smol,
+		Stream:    &stream,
+		KeepAlive: &api.Duration{Duration: 10 * time.Second},
+		Options: map[string]any{
+			"num_ctx": 16384,
+		},
+		Messages: []api.Message{
+			{
+				Role:    "user",
+				Content: rainbowPrompt,
+			},
+		},
+	}
+
+	softTimeout, hardTimeout := getTimeouts(t)
+	ctx, cancel := context.WithTimeout(context.Background(), hardTimeout)
+	defer cancel()
+	client, _, cleanup := InitServerConnection(ctx, t)
+	defer cleanup()
+
+	// Get the server running (if applicable) warm the model up with a single initial request
+	slog.Info("loading", "model", req.Model)
+	err := client.Generate(ctx,
+		&api.GenerateRequest{Model: req.Model, KeepAlive: &api.Duration{Duration: 10 * time.Second}, Options: req.Options},
+		func(response api.GenerateResponse) error { return nil },
+	)
+	if err != nil {
+		t.Fatalf("failed to load model %s: %s", req.Model, err)
+	}
+
+	assistant := DoChat(ctx, t, client, req, rainbowExpected, 30*time.Second, 20*time.Second)
+
+	for i := 0; i < len(rainbowFollowups); i++ {
+		if time.Now().Sub(started) > softTimeout {
+			slog.Info("exceeded soft timeout, winding down test")
+			return
+		}
+		req.Messages = append(req.Messages,
+			*assistant,
+			api.Message{Role: "user", Content: rainbowFollowups[i]},
+		)
+
+		assistant = DoChat(ctx, t, client, req, rainbowExpected, 30*time.Second, 20*time.Second)
+		if assistant == nil {
+			t.Fatalf("didn't get an assistant response for context")
+		}
+	}
+}

integration/library_models_test.go

@@ -4,7 +4,9 @@ package integration
 
 import (
 	"context"
+	"fmt"
 	"log/slog"
+	"os"
 	"testing"
 	"time"
 
@@ -20,6 +22,7 @@ func TestLibraryModelsGenerate(t *testing.T) {
 	defer cancel()
 	client, _, cleanup := InitServerConnection(ctx, t)
 	defer cleanup()
+	targetArch := os.Getenv("OLLAMA_TEST_ARCHITECTURE")
 
 	chatModels := libraryChatModels
 	for _, model := range chatModels {
@@ -30,16 +33,26 @@ func TestLibraryModelsGenerate(t *testing.T) {
 			if err := PullIfMissing(ctx, client, model); err != nil {
 				t.Fatalf("pull failed %s", err)
 			}
+			if targetArch != "" {
+				resp, err := client.Show(ctx, &api.ShowRequest{Name: model})
+				if err != nil {
+					t.Fatalf("unable to show model: %s", err)
+				}
+				arch := resp.ModelInfo["general.architecture"].(string)
+				if arch != targetArch {
+					t.Skip(fmt.Sprintf("Skipping %s architecture %s != %s", model, arch, targetArch))
+				}
+			}
 			req := api.GenerateRequest{
 				Model:     model,
-				Prompt:    "why is the sky blue?",
+				Prompt:    blueSkyPrompt,
 				KeepAlive: &api.Duration{Duration: 10 * time.Second},
 				Options: map[string]interface{}{
 					"temperature": 0.1,
 					"seed":        123,
 				},
 			}
-			anyResp := []string{"rayleigh", "scatter", "atmosphere", "nitrogen", "oxygen", "wavelength"}
+			anyResp := blueSkyExpected
 			// Special cases
 			if model == "duckdb-nsql" {
 				anyResp = []string{"select", "from"}

integration/model_arch_test.go

@@ -68,14 +68,13 @@ func TestModelsGenerate(t *testing.T) {
 			// TODO - fiddle with context size
 			req := api.GenerateRequest{
 				Model:  model,
-				Prompt: "why is the sky blue?",
+				Prompt: blueSkyPrompt,
 				Options: map[string]interface{}{
 					"temperature": 0,
 					"seed":        123,
 				},
 			}
-			anyResp := []string{"rayleigh", "scattering", "atmosphere", "nitrogen", "oxygen"}
+			DoGenerate(ctx, t, client, req, blueSkyExpected, 120*time.Second, 30*time.Second)
-			DoGenerate(ctx, t, client, req, anyResp, 120*time.Second, 30*time.Second)
 		})
 	}
 }

integration/model_perf_test.go

@@ -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
+	targetArch := os.Getenv("OLLAMA_TEST_ARCHITECTURE")
-	if s := os.Getenv("OLLAMA_NEW_ENGINE"); s != "" {
-		chatModels = ollamaEngineChatModels
-	} else {
-		chatModels = append(ollamaEngineChatModels, llamaRunnerChatModels...)
-	}
 
 	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"}},
+					{blueSkyPrompt, blueSkyExpected},
-					{maxPrompt, []string{"shakespeare", "oppression", "sorrows", "gutenberg", "child", "license", "sonnet", "melancholy"}},
+					{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),

integration/quantization_test.go

@@ -76,7 +76,7 @@ func TestQuantization(t *testing.T) {
 				stream := true
 				genReq := api.GenerateRequest{
 					Model:     newName,
-					Prompt:    "why is the sky blue?",
+					Prompt:    blueSkyPrompt,
 					KeepAlive: &api.Duration{Duration: 3 * time.Second},
 					Options: map[string]any{
 						"seed":        42,
@@ -88,14 +88,13 @@ func TestQuantization(t *testing.T) {
 
 				// Some smaller quantizations can cause models to have poor quality
 				// or get stuck in repetition loops, so we stop as soon as we have any matches
-				anyResp := []string{"rayleigh", "scattering", "day", "sun", "moon", "color", "nitrogen", "oxygen"}
 				reqCtx, reqCancel := context.WithCancel(ctx)
 				atLeastOne := false
 				var buf bytes.Buffer
 				genfn := func(response api.GenerateResponse) error {
 					buf.Write([]byte(response.Response))
 					fullResp := strings.ToLower(buf.String())
-					for _, resp := range anyResp {
+					for _, resp := range blueSkyExpected {
 						if strings.Contains(fullResp, resp) {
 							atLeastOne = true
 							t.Log(fullResp)

integration/utils_test.go

@@ -256,13 +256,29 @@ var (
 		"snowflake-arctic-embed",
 		"snowflake-arctic-embed2",
 	}
+
+	blueSkyPrompt   = "why is the sky blue? Be brief but factual in your reply"
+	blueSkyExpected = []string{"rayleigh", "scatter", "atmosphere", "nitrogen", "oxygen", "wavelength", "interact"}
+
+	rainbowPrompt    = "how do rainbows form? Be brief but factual in your reply"
+	rainbowFollowups = []string{
+		"Explain the physics involved in them.  Be breif in your reply",
+		"Explain the chemistry involved in them.  Be breif in your reply",
+		"Explain the quantum mechanics involved in them. Be breif in your reply",
+		"What are common myths related to them? Be brief in your reply",
+		"What are common fairytales related to them? Be brief in your reply",
+		"Can they form if there is no rain?  Be breif in your reply",
+		"Can they form if there are no clouds?  Be breif in your reply",
+		"Do they happen on other planets? Be brief in your reply",
+	}
+	rainbowExpected = []string{"water", "droplet", "mist", "glow", "refracted", "reflect", "color", "spectrum", "frequency", "end", "gold", "fortune", "blessing", "prosperity"}
 )
 
 func init() {
 	lifecycle.InitLogging()
-	custom := os.Getenv("OLLAMA_TEST_SMOL_MODEL")
+	custom := os.Getenv("OLLAMA_TEST_DEFAULT_MODEL")
 	if custom != "" {
-		slog.Info("setting smol test model to " + custom)
+		slog.Info("setting default test model to " + custom)
 		smol = custom
 	}
 }
@@ -577,11 +593,11 @@ func GenerateRequests() ([]api.GenerateRequest, [][]string) {
 			},
 		},
 		[][]string{
-			{"sunlight", "scattering", "interact", "color", "surface", "depth", "red", "orange", "yellow", "absorbs", "wavelength"},
+			{"sunlight", "scatter", "interact", "color", "surface", "depth", "red", "orange", "yellow", "absorb", "wavelength", "water", "molecule"},
-			{"soil", "organic", "earth", "black", "tan", "chemical", "processes", "pigments", "particles", "iron oxide", "rust", "air", "water", "mixture", "mixing"},
+			{"soil", "organic", "earth", "black", "tan", "chemical", "processes", "pigment", "particle", "iron oxide", "rust", "air", "water", "wet", "mixture", "mixing", "mineral", "element", "decomposed", "matter", "wavelength"},
-			{"water", "droplet", "refracted", "reflect", "color", "spectrum"},
+			{"water", "droplet", "refract", "reflect", "color", "spectrum", "raindrop"},
 			{"fourth", "july", "declaration", "independence"},
-			{"nitrogen", "oxygen", "carbon", "dioxide", "water", "vapor"},
+			{"nitrogen", "oxygen", "carbon", "dioxide", "water", "vapor", "fluid", "particles", "gas"},
 		}
 }