Add yet more debugging information to LLM test.

The current failures are consistent with a timeout when generating
the first prompt, so increase a timeout that a past change forgot to
increase for the first test.

Otherwise, include server-side logs into error output to hopefully shed some
light on what is happening.

PiperOrigin-RevId: 599002718

test/gpu/ollama/ollama.go

@@ -356,7 +356,17 @@ type ConversationContext []int
 func (llm *Ollama) Prompt(ctx context.Context, prompt *Prompt) (*Response, error) {
 	resp, err := jsonPost[PromptJSON, ResponseJSON](ctx, llm, "/api/generate", prompt.json())
 	if err != nil {
-		return nil, err
+		if ctx.Err() != nil {
+			return nil, fmt.Errorf("%w (+ context err: %v)", err, ctx.Err())
+		}
+		serverLogs, logsErr := llm.container.Logs(ctx)
+		if logsErr != nil {
+			return nil, fmt.Errorf("%w (could not get server logs: %v)", err, logsErr)
+		}
+		if serverLogs != "" {
+			return nil, fmt.Errorf("%w; ollama server logs:\n%v\n(end of ollama server logs)", err, serverLogs)
+		}
+		return nil, fmt.Errorf("%w (server logs are empty)", err)
 	}
 	return &Response{data: resp}, nil
 }

test/gpu/textgen_test.go

@@ -91,7 +91,7 @@ func TestLLM(t *testing.T) {
 				    "Hello World".
 			`,
 		}
-		promptCtx, promptCancel := context.WithTimeout(ctx, time.Minute)
+		promptCtx, promptCancel := context.WithTimeout(ctx, 3*time.Minute)
 		response, err := llm.PromptUntil(promptCtx, &prompt, func(prompt *ollama.Prompt, response *ollama.Response) (*ollama.Prompt, error) {
 			defer prompt.Model.RaiseTemperature()
 			text := strings.TrimSpace(response.Text())