docs(gpu-ai): added example demonstrating GPU usage with Ollama AI model

examples/rental-model/advanced/gpu-ai.ts

@@ -0,0 +1,101 @@
+import { GolemNetwork, MarketOrderSpec, waitFor } from "@golem-sdk/golem-js";
+import { pinoPrettyLogger } from "@golem-sdk/pino-logger";
+import chalk from "chalk";
+
+(async () => {
+  const glm = new GolemNetwork({
+    logger: pinoPrettyLogger({
+      level: "info",
+    }),
+  });
+
+  try {
+    await glm.connect();
+    const network = await glm.createNetwork({ ip: "192.168.0.0/24" });
+
+    const order: MarketOrderSpec = {
+      demand: {
+        workload: {
+          // ollama with qwen2:0.5b
+          imageHash: "23ac8d8f54623ad414d70392e4e3b96da177911b0143339819ec1433",
+          minMemGib: 8,
+          capabilities: ["!exp:gpu", "vpn"],
+          runtime: { name: "vm-nvidia" },
+        },
+      },
+      market: {
+        rentHours: 0.5,
+        pricing: {
+          model: "linear",
+          maxStartPrice: 0.0,
+          maxCpuPerHourPrice: 0.0,
+          maxEnvPerHourPrice: 2.0,
+        },
+      },
+      network,
+    };
+
+    const abortController = new AbortController();
+    const signal = abortController.signal;
+    process.on("SIGINT", () => abortController.abort("Process interrupted at user request"));
+
+    const rental = await glm.oneOf({ order, signalOrTimeout: signal });
+    const exe = await rental.getExeUnit(signal);
+
+    const PORT_ON_PROVIDER = 11434;
+    const PORT_ON_REQUESTOR = 11434;
+    let isServerReady = false;
+
+    console.log(`Starting Ollama on provider ${exe.provider.name}...`);
+
+    const server = await exe.runAndStream("sleep 1 && /usr/bin/ollama serve", {
+      signalOrTimeout: signal,
+    });
+
+    server.stdout.subscribe((data) => console.log(chalk.yellow(data?.toString().trim())));
+    server.stderr.subscribe((data) => {
+      console.log(chalk.yellow(data?.toString().trim()));
+      if (data?.toString().includes("Listening on [::]:11434")) {
+        isServerReady = true;
+      }
+    });
+
+    await waitFor(() => isServerReady, { abortSignal: signal });
+
+    const proxy = exe.createTcpProxy(PORT_ON_PROVIDER);
+    proxy.events.on("error", (error) => console.error("TcpProxy reported an error:", error));
+
+    await proxy.listen(PORT_ON_REQUESTOR);
+    console.log(
+      `Server Proxy listen at http://localhost:${PORT_ON_REQUESTOR}\n` +
+        "Now you can talk to the model, for example using the command:\n\n" +
+        chalk.magenta(
+          `curl http://localhost:11434/v1/chat/completions -d '{ "model": "qwen2:0.5b", "messages": [ { "role": "user", "content": "What is Golem?" } ]}'\n`,
+        ),
+    );
+
+    let isClosing = false;
+    const stopServer = async () => {
+      if (isClosing) {
+        console.log("Already closing, ignoring subsequent shutdown request. Process PID: %d", process.pid);
+        return;
+      }
+      isClosing = true;
+      console.log("Shutting down gracefully");
+      await proxy.close();
+      console.log("Shutdown routine completed");
+    };
+
+    abortController.signal.addEventListener("abort", () =>
+      stopServer()
+        .catch((err) => console.error("Something went wrong while stopping the server", err))
+        .finally(() => rental.stopAndFinalize()),
+    );
+
+    await waitFor(() => server.isFinished(), { abortSignal: AbortSignal.timeout(120_000) });
+  } catch (err) {
+    console.error("Failed to run the example", err);
+  } finally {
+    await glm.disconnect();
+  }
+})().catch(console.error);