@qvac/sdk 0.8.0 → 1.0.0

package/dist/client/rpc/rpc-client.js

@@ -1,14 +1,8 @@
-import { requestSchema, responseSchema, PROFILING_TRAILER_KEY, } from "../../schemas/index.js";
+import { requestSchema, responseSchema, } from "../../schemas/index.js";
 import { RPCError } from "./rpc-error.js";
 import { withTimeout, withTimeoutStream } from "../../utils/withTimeout.js";
-import { getClientLogger, summarizeRequest } from "../../logging/index.js";
-import { getRPC, close as closeRPC } from "#rpc";
-import { nowMs, shouldProfile, shouldIncludeServerBreakdown, generateId as createProfileId, createProfilingMeta, createProfilingDisabledMeta, injectProfilingMetaIntoObject, extractProfilingMeta, stripProfilingMeta, recordFailure, } from "../../profiling/index.js";
-import { createClientTimings, createClientStreamTimings, recordClientEvents, recordClientStreamEvents, cacheConnectionTime, flushConnectionTime, resetConnectionTracking, } from "./profiling/index.js";
-const logger = getClientLogger();
 let rpcInstance = null;
 let commandCounter = 0;
-let firstConnectionPending = true;
 function getNextCommandId() {
     commandCounter = (commandCounter + 1) % Number.MAX_SAFE_INTEGER;
     return commandCounter;
@@ -18,115 +12,67 @@ function checkAndThrowError(response) {
         throw new RPCError(response);
     }
 }
-async function getRPCInstance() {
+async function getRPC() {
     if (rpcInstance)
-        return { rpc: await rpcInstance };
-    const connectionStart = firstConnectionPending ? nowMs() : null;
-    rpcInstance = getRPC();
-    const rpc = await rpcInstance;
-    if (connectionStart !== null && firstConnectionPending) {
-        firstConnectionPending = false;
-        return { rpc, connectionMs: nowMs() - connectionStart };
+        return rpcInstance;
+    const isRN = typeof navigator !== "undefined" && navigator.product === "ReactNative";
+    // Detect Bare runtime - Bare exposes global Bare object
+    // TODO we can use something like https://www.npmjs.com/package/which-runtime at some point
+    const isBare = typeof globalThis !== "undefined" && "Bare" in globalThis;
+    // TODO Lots of type unsafe hacks to make this work cross platform
+    if (isRN) {
+        console.log("React Native runtime detected");
+        // Only import expo client for React Native - Metro won't bundle the Node client
+        const expoRPCClientPath = "./expo-rpc-client";
+        const mod = (await import(expoRPCClientPath));
+        rpcInstance = await (typeof mod === "function" ? mod() : mod.getRPC());
+        return rpcInstance;
     }
-    return { rpc };
-}
-async function prepareRPCContext(requestType, perCallProfiling, rpc) {
-    const rpcResult = rpc ? { rpc } : await getRPCInstance();
-    const profilingEnabled = shouldProfile(requestType, perCallProfiling);
-    const signalDisable = perCallProfiling?.enabled === false;
-    if (rpcResult.connectionMs !== undefined) {
-        cacheConnectionTime(rpcResult.connectionMs);
-    }
-    if (profilingEnabled) {
-        flushConnectionTime();
+    else if (isBare) {
+        console.log("Bare runtime detected");
+        // Use dynamic import for Bare client
+        const mod = await import("./bare-client.js");
+        rpcInstance = mod.getRPC();
+        return rpcInstance;
     }
-    return { rpc: rpcResult.rpc, profilingEnabled, signalDisable };
-}
-export async function send(request, options, rpc) {
-    const ctx = await prepareRPCContext(request.type, options?.profiling, rpc);
-    if (!ctx.profilingEnabled) {
-        return sendBase(request, ctx.rpc, options, ctx.signalDisable);
+    else {
+        console.log("Node runtime detected");
+        // Use dynamic import for Node client to prevent Metro from bundling it
+        const mod = await import("./node-rpc-client.js");
+        rpcInstance = mod.getRPC();
+        return rpcInstance;
     }
-    return sendProfiled(request, ctx.rpc, options);
 }
-async function sendBase(request, rpc, options, signalDisable = false) {
+export async function send(request, rpc, options) {
     const parsedRequest = requestSchema.parse(request);
-    const req = rpc.request(getNextCommandId());
-    logger.debug("RPC Client sending:", summarizeRequest(request));
-    const payloadObj = signalDisable
-        ? injectProfilingMetaIntoObject(parsedRequest, createProfilingDisabledMeta())
-        : parsedRequest;
-    const payload = JSON.stringify(payloadObj);
+    const rpcInstance = rpc || (await getRPC());
+    const req = rpcInstance.request(getNextCommandId());
+    // TODO temp for debugging, find a better solution to exclude large payloads
+    // Log request type and basic info, avoid logging large payloads
+    if (request.type === "transcribeStream") {
+        console.log("RPC Client sending:", {
+            type: request.type,
+            modelId: request.modelId,
+            audioChunkType: request.audioChunk.type,
+            audioChunkSize: request.audioChunk.value.length,
+        });
+    }
+    else {
+        console.log("RPC Client sending:", request);
+    }
+    const payload = JSON.stringify(parsedRequest);
     req.send(payload, "utf-8");
     const response = await withTimeout(req.reply("utf-8"), options?.timeout);
     const resPayload = responseSchema.parse(JSON.parse(response?.toString() || "{}"));
-    logger.debug("ResPayload", { type: resPayload.type });
+    console.log("ResPayload", { type: resPayload.type });
     checkAndThrowError(resPayload);
     return resPayload;
 }
-async function sendProfiled(request, rpc, options) {
-    const requestType = request.type;
-    const profileId = createProfileId();
-    const includeServer = shouldIncludeServerBreakdown(options?.profiling);
-    const timings = createClientTimings(profileId, requestType);
-    try {
-        const zodStart = nowMs();
-        const parsedRequest = requestSchema.parse(request);
-        timings.requestZodValidationMs = nowMs() - zodStart;
-        const req = rpc.request(getNextCommandId());
-        logger.debug("RPC Client sending:", summarizeRequest(request));
-        const profilingMeta = createProfilingMeta(profileId, includeServer);
-        const requestWithMeta = injectProfilingMetaIntoObject(parsedRequest, profilingMeta);
-        const stringifyStart = nowMs();
-        const payload = JSON.stringify(requestWithMeta);
-        timings.requestStringifyMs = nowMs() - stringifyStart;
-        timings.sendStart = nowMs();
-        req.send(payload, "utf-8");
-        const response = await withTimeout(req.reply("utf-8"), options?.timeout);
-        timings.firstResponseAt = nowMs();
-        const parseStart = nowMs();
-        const rawParsed = JSON.parse(response?.toString() || "{}");
-        timings.responseJsonParseMs = nowMs() - parseStart;
-        const responseMeta = extractProfilingMeta(rawParsed);
-        const cleanPayload = stripProfilingMeta(rawParsed);
-        const zodValidateStart = nowMs();
-        const resPayload = responseSchema.parse(cleanPayload);
-        timings.responseZodValidationMs = nowMs() - zodValidateStart;
-        timings.requestEnd = nowMs();
-        logger.debug("ResPayload", { type: resPayload.type });
-        recordClientEvents(timings, responseMeta);
-        checkAndThrowError(resPayload);
-        return resPayload;
-    }
-    catch (error) {
-        if (timings.requestEnd === undefined) {
-            const base = {
-                ts: nowMs(),
-                op: timings.requestType,
-                kind: "rpc",
-                profileId: timings.profileId,
-            };
-            recordFailure(base, timings.requestStart, error);
-        }
-        throw error;
-    }
-}
-export async function* stream(request, options = {}, rpc) {
-    const ctx = await prepareRPCContext(request.type, options?.profiling, rpc);
-    if (!ctx.profilingEnabled) {
-        yield* streamBase(request, ctx.rpc, options, ctx.signalDisable);
-        return;
-    }
-    yield* streamProfiled(request, ctx.rpc, options);
-}
-async function* streamBase(request, rpc, options = {}, signalDisable = false) {
+export async function* stream(request, rpc, options = {}) {
     const parsedRequest = requestSchema.parse(request);
-    const req = rpc.request(getNextCommandId());
-    logger.debug("RPC Client streaming:", summarizeRequest(request));
-    const payloadObj = signalDisable
-        ? injectProfilingMetaIntoObject(parsedRequest, createProfilingDisabledMeta())
-        : parsedRequest;
-    req.send(JSON.stringify(payloadObj), "utf-8");
+    const rpcInstance = rpc || (await getRPC());
+    const req = rpcInstance.request(getNextCommandId());
+    req.send(JSON.stringify(parsedRequest), "utf-8");
     const responseStream = req.createResponseStream({ encoding: "utf-8" });
     let buffer = "";
     async function* processStream() {
@@ -149,84 +95,3 @@ async function* streamBase(request, rpc, options = {}, signalDisable = false) {
         }
     }
 }
-async function* streamProfiled(request, rpc, options = {}) {
-    const requestType = request.type;
-    const profileId = createProfileId();
-    const includeServer = shouldIncludeServerBreakdown(options?.profiling);
-    const timings = createClientStreamTimings(profileId, requestType);
-    let profilingMeta = undefined;
-    try {
-        const zodStart = nowMs();
-        const parsedRequest = requestSchema.parse(request);
-        timings.requestZodValidationMs = nowMs() - zodStart;
-        const req = rpc.request(getNextCommandId());
-        logger.debug("RPC Client streaming:", summarizeRequest(request));
-        const requestMeta = createProfilingMeta(profileId, includeServer);
-        const requestWithMeta = injectProfilingMetaIntoObject(parsedRequest, requestMeta);
-        const stringifyStart = nowMs();
-        const payload = JSON.stringify(requestWithMeta);
-        timings.requestStringifyMs = nowMs() - stringifyStart;
-        timings.sendStart = nowMs();
-        req.send(payload, "utf-8");
-        const responseStream = req.createResponseStream({ encoding: "utf-8" });
-        let buffer = "";
-        async function* processStream() {
-            for await (const chunk of responseStream) {
-                yield chunk;
-            }
-        }
-        const streamWithTimeout = withTimeoutStream(processStream(), options?.timeout);
-        for await (const chunk of streamWithTimeout) {
-            const chunkTime = nowMs();
-            if (timings.firstChunkAt === undefined) {
-                timings.firstChunkAt = chunkTime;
-            }
-            timings.lastChunkAt = chunkTime;
-            buffer += chunk.toString();
-            const lines = buffer.split("\n");
-            buffer = lines.pop() || "";
-            for (const line of lines) {
-                if (line.trim()) {
-                    const rawParsed = JSON.parse(line);
-                    const chunkMeta = extractProfilingMeta(rawParsed);
-                    if (chunkMeta) {
-                        profilingMeta = chunkMeta;
-                    }
-                    if (rawParsed[PROFILING_TRAILER_KEY] === true)
-                        continue;
-                    const cleanPayload = stripProfilingMeta(rawParsed);
-                    const response = responseSchema.parse(cleanPayload);
-                    timings.chunkCount++;
-                    checkAndThrowError(response);
-                    yield response;
-                }
-            }
-        }
-    }
-    catch (error) {
-        if (timings.requestEnd === undefined) {
-            const base = {
-                ts: nowMs(),
-                op: timings.requestType,
-                kind: "rpc",
-                profileId: timings.profileId,
-            };
-            recordFailure(base, timings.requestStart, error);
-        }
-        throw error;
-    }
-    finally {
-        if (timings.chunkCount > 0) {
-            timings.requestEnd = nowMs();
-            recordClientStreamEvents(timings, profilingMeta);
-        }
-    }
-}
-export async function close() {
-    if (!rpcInstance)
-        return;
-    rpcInstance = null;
-    firstConnectionPending = true;
-    resetConnectionTracking();
-    await closeRPC();
-}

package/dist/constants/audio.d.ts

@@ -1,2 +1,3 @@
-export { FORMATS_NEEDING_DECODE, SUPPORTED_AUDIO_FORMATS, } from "@qvac/decoder-audio/constants";
+export declare const FORMATS_NEEDING_DECODE: readonly [".mp3", ".m4a", ".ogg", ".flac", ".aac", ".wav"];
+export declare const SUPPORTED_AUDIO_FORMATS: readonly [".mp3", ".m4a", ".ogg", ".wav", ".flac", ".aac", ".raw"];
 //# sourceMappingURL=audio.d.ts.map

package/dist/constants/audio.js

@@ -1 +1,17 @@
-export { FORMATS_NEEDING_DECODE, SUPPORTED_AUDIO_FORMATS, } from "@qvac/decoder-audio/constants.js";
+export const FORMATS_NEEDING_DECODE = [
+    ".mp3",
+    ".m4a",
+    ".ogg",
+    ".flac",
+    ".aac",
+    ".wav",
+];
+export const SUPPORTED_AUDIO_FORMATS = [
+    ".mp3",
+    ".m4a",
+    ".ogg",
+    ".wav",
+    ".flac",
+    ".aac",
+    ".raw",
+];

package/dist/constants/index.d.ts

@@ -1,3 +1,2 @@
 export * from "./audio";
-export * from "./registry";
 //# sourceMappingURL=index.d.ts.map

package/dist/constants/index.js

@@ -1,2 +1 @@
 export * from "./audio.js";
-export * from "./registry.js";

package/dist/examples/delegated-inference/consumer.js

@@ -1,4 +1,4 @@
-import { completion, LLAMA_3_2_1B_INST_Q4_0, loadModel, close, } from "@qvac/sdk";
+import { completion, LLAMA_3_2_1B_INST_Q4_0, loadModel } from "@qvac/sdk";
 // Topic hex (required)
 const topicHex = process.argv[2];
 if (!topicHex) {
@@ -32,7 +32,6 @@ try {
             providerPublicKey,
             timeout: 5_000, // Optional: 5 second timeout for delegated requests
             fallbackToLocal: true, // Optional: Fall back to local inference if delegation fails
-            // forceNewConnection: true, // Optional: Force a new connection instead of reusing cached one
         },
         onProgress: (progress) => {
             console.log(`📊 Download progress: ${progress.percentage.toFixed(1)}% (${progress.downloaded}/${progress.total} bytes)`);
@@ -49,7 +48,7 @@ try {
     }
     console.log("🔍 Stats:", await response.stats);
     console.log("\n🎯 Delegation infrastructure working! Server correctly detected and routed the delegated request.");
-    void close();
+    process.exit(0);
 }
 catch (error) {
     console.error("❌ Error:", error);

package/dist/examples/download-with-cancel.js

@@ -1,4 +1,4 @@
-import { cancel, close, downloadAsset, LLAMA_3_2_1B_INST_Q4_0, } from "@qvac/sdk";
+import { downloadAsset, cancel, LLAMA_3_2_1B_INST_Q4_0 } from "@qvac/sdk";
 console.log(`🚀 Starting download with pause/resume example`);
 console.log(`\n💡 Press Ctrl+C to pause the download (it will resume on restart)\n`);
 let modelId;
@@ -31,12 +31,13 @@ try {
     });
     console.log(`\n✅ Model downloaded successfully! Model ID: ${modelId}`);
     console.log("🎯 Download completed without interruption");
-    void close();
+    // Clean up
+    process.exit(0);
 }
 catch (error) {
     if (error instanceof Error && error.message.includes("cancelled")) {
         console.log("✅ Download was successfully cancelled");
-        void close();
+        process.exit(0);
     }
     else {
         console.error("❌ Error:", error);

package/dist/examples/embed-hyperdrive.js

@@ -1,11 +1,4 @@
 import { embed, GTE_LARGE_FP16, loadModel, unloadModel } from "@qvac/sdk";
-function cosineSimilarity(vecA, vecB) {
-    let dotProduct = 0;
-    for (let i = 0; i < vecA.length; i++) {
-        dotProduct += vecA[i] * vecB[i];
-    }
-    return dotProduct;
-}
 try {
     const modelId = await loadModel({
         modelSrc: GTE_LARGE_FP16,
@@ -13,40 +6,12 @@ try {
         onProgress: (progress) => {
             console.log(progress);
         },
-        modelConfig: {
-            gpuLayers: 99,
-            device: "gpu",
-        },
     });
-    console.log("\n📝 Example 1: Single Text Embedding");
-    console.log("=".repeat(50));
-    const singleEmbedding = await embed({ modelId, text: "Hello, world!" });
-    console.log("Input: 'Hello, world!'");
-    console.log("Embedding dimensions:", singleEmbedding.length);
-    console.log("First 10 values:", singleEmbedding.slice(0, 10));
-    console.log("\n📝 Example 2: Batch Text Embeddings");
-    console.log("=".repeat(50));
-    const texts = [
-        "The quick brown fox jumps over the lazy dog",
-        "A fast auburn fox leaps over a sleepy canine",
-        "Python is a programming language",
-    ];
-    const batchEmbeddings = await embed({ modelId, text: texts });
-    console.log("Input: Array of", texts.length, "texts");
-    console.log("Output: Array of", batchEmbeddings.length, "embeddings");
-    const [emb1, emb2, emb3] = batchEmbeddings;
-    if (!emb1 || !emb2 || !emb3) {
-        throw new Error("Expected 3 embeddings");
-    }
-    console.log("Each embedding dimensions:", emb1.length);
-    console.log("\n🔍 Similarity Analysis");
-    console.log("=".repeat(50));
-    const similarity1 = cosineSimilarity(emb1, emb2);
-    const similarity2 = cosineSimilarity(emb1, emb3);
-    console.log("Similarity between texts 1 and 2 (similar meaning):", similarity1.toFixed(4));
-    console.log("Similarity between texts 1 and 3 (different topics):", similarity2.toFixed(4));
-    console.log("\n💡 Higher values indicate more similar meanings");
+    const embedding = await embed({ modelId, text: "Hello, world!" });
+    console.log("Embedding (first 20 elements)", embedding.slice(0, 20));
+    console.log("Embeddings length", embedding.length);
     await unloadModel({ modelId, clearStorage: false });
+    process.exit(0);
 }
 catch (error) {
     console.error("❌ Error:", error);

package/dist/examples/kv-cache-cleanup.js

@@ -9,7 +9,7 @@ try {
         modelConfig: {
             device: "gpu",
             ctx_size: 2048,
-            verbosity: VERBOSITY.ERROR,
+            verbosity: VERBOSITY.SILENT,
         },
     });
     // Create some cached sessions
@@ -80,6 +80,7 @@ try {
     console.log("   - Privacy → Clear conversation history");
     console.log("   - App reset → Delete all caches with { all: true }");
     await unloadModel({ modelId, clearStorage: false });
+    process.exit(0);
 }
 catch (error) {
     console.error("❌ Error:", error);

package/dist/examples/kv-cache-custom-key.js

@@ -14,7 +14,7 @@ try {
         modelConfig: {
             device: "gpu",
             ctx_size: 2048,
-            verbosity: VERBOSITY.DEBUG,
+            verbosity: VERBOSITY.SILENT,
         },
     });
     // Prompt 1 with cache key "session-a" (first time - no cache exists)
@@ -70,21 +70,6 @@ try {
     }
     const stats3 = await result3.stats;
     console.log(`\n   Stats: TTFT=${stats3?.timeToFirstToken}ms, CacheTokens=${stats3?.cacheTokens}\n`);
-    // Prompt 4: Go back to session-a to trigger flush of session-b
-    console.log('\n📝 Prompt 4: "What is 5+5?" with cache key "session-a" (BACK TO FIRST)');
-    console.log("   Status: This should trigger session-b to be flushed to disk\n");
-    const result4 = completion({
-        modelId,
-        history: [systemPromptMessage, { role: "user", content: "What is 5+5?" }],
-        stream: true,
-        kvCache: "session-a", // Back to session-a!
-    });
-    console.log("   Response: ");
-    for await (const token of result4.tokenStream) {
-        process.stdout.write(token);
-    }
-    const stats4 = await result4.stats;
-    console.log(`\n   Stats: TTFT=${stats4?.timeToFirstToken}ms, CacheTokens=${stats4?.cacheTokens}\n`);
     // Summary
     console.log("\n✅ Summary:");
     console.log('   - Prompts 1 & 2 share cache key "session-a" → same cache file');
@@ -94,6 +79,7 @@ try {
     console.log("   - Different cache keys provide complete isolation between sessions");
     await deleteCache({ all: true });
     await unloadModel({ modelId, clearStorage: false });
+    process.exit(0);
 }
 catch (error) {
     console.error("❌ Error:", error);

package/dist/examples/kv-cache-example.js

@@ -7,7 +7,7 @@ try {
         modelConfig: {
             device: "gpu",
             ctx_size: 2048,
-            verbosity: VERBOSITY.ERROR,
+            verbosity: VERBOSITY.SILENT,
         },
     });
     console.log("🧠 Testing KV Cache functionality...\n");
@@ -66,6 +66,7 @@ try {
     console.log(`\n⏱️  Non-cached completion stats: ${JSON.stringify(stats3)}\n`);
     console.log("✅ KV Cache test completed!");
     await unloadModel({ modelId, clearStorage: false });
+    process.exit(0);
 }
 catch (error) {
     console.error("❌ Error:", error);

package/dist/examples/llamacpp-file-logging.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=llamacpp-file-logging.d.ts.map

package/dist/examples/{logging-file-transport.js → llamacpp-file-logging.js}

@@ -1,6 +1,8 @@
-import { getLogger, loadModel, completion, unloadModel, LLAMA_3_2_1B_INST_Q4_0, VERBOSITY, } from "@qvac/sdk";
+import { getLogger } from "../logging/index.js";
+import { loadModel, completion, unloadModel } from "../client/api/index.js";
 import { writeFileSync, appendFileSync } from "fs";
 import { join } from "path";
+import { LLAMA_3_2_1B_INST_Q4_0 } from "../models/hyperdrive/index.js";
 try {
     const logFile = join(process.cwd(), "model-logs.txt");
     // Clear previous logs
@@ -8,7 +10,6 @@ try {
     // Create logger with file transport
     const logger = getLogger("llm-file-demo", {
         level: "debug",
-        enableConsole: false, // Disable console output (logs go to file only). Set to true for both console + file
         transports: [
             // File transport - append logs to file
             (level, namespace, message) => {
@@ -16,6 +17,10 @@ try {
                 const logLine = `[${timestamp}] ${level.toUpperCase()} ${namespace}: ${message}\n`;
                 appendFileSync(logFile, logLine);
             },
+            // Console transport for immediate feedback
+            (level, namespace, message) => {
+                console.log(`🔍 [${level.toUpperCase()}] ${namespace}: ${message}`);
+            },
         ],
     });
     logger.info("Starting model demo with file logging");
@@ -26,7 +31,6 @@ try {
         modelConfig: {
             ctx_size: 2048,
             temp: 0.7,
-            verbosity: VERBOSITY.DEBUG,
         },
         logger, // All model logs will be captured
     });
@@ -50,6 +54,7 @@ try {
     // Show user where logs were saved
     console.log(`\n📄 All logs (including model internals) saved to: ${logFile}`);
     console.log("You can inspect the file to see detailed model operation logs!");
+    process.exit(0);
 }
 catch (error) {
     console.error("❌ Error:", error);

package/dist/examples/llamacpp-filesystem.js

@@ -1,4 +1,4 @@
-import { completion, loadModel, unloadModel } from "@qvac/sdk";
+import { completion, loadModel } from "@qvac/sdk";
 // Get GGUF file path from command line arguments
 const ggufPath = process.argv[2];
 if (!ggufPath) {
@@ -32,7 +32,7 @@ try {
     const stats = await result.stats;
     console.log("\n📊 Performance Stats:", stats);
     console.log("\n\n🎉 Completed!");
-    await unloadModel({ modelId, clearStorage: false });
+    process.exit(0);
 }
 catch (error) {
     console.error("❌ Error:", error);

package/dist/examples/llamacpp-http.js

@@ -33,6 +33,7 @@ try {
     console.log("\n📊 Performance Stats:", stats);
     console.log("\n\n🎉 Completed!");
     await unloadModel({ modelId, clearStorage: false });
+    process.exit(0);
 }
 catch (error) {
     console.error("❌ Error:", error);

package/dist/examples/llamacpp-hyperdrive.js

@@ -14,7 +14,7 @@ try {
         modelConfig: {
             device: "gpu",
             ctx_size: 2048,
-            verbosity: VERBOSITY.ERROR,
+            verbosity: VERBOSITY.SILENT,
         },
     });
     const history = [
@@ -31,6 +31,7 @@ try {
     console.log("\n📊 Performance Stats:", stats);
     // Change `clearStorage: true` to delete cached model files
     await unloadModel({ modelId, clearStorage: false });
+    process.exit(0);
 }
 catch (error) {
     console.error("❌ Error:", error);

package/dist/examples/llamacpp-multimodal.js

@@ -1,4 +1,4 @@
-import { completion, loadModel, SMOLVLM2_500M_MULTIMODAL_Q8_0, MMPROJ_SMOLVLM2_500M_MULTIMODAL_Q8_0, unloadModel, } from "@qvac/sdk";
+import { completion, loadModel, SMOLVLM2_2_500M_MULTIMODAL_Q8_0, MMPROJ_SMOLVLM2_2_500M_MULTIMODAL_Q8_0, unloadModel, } from "@qvac/sdk";
 if (process.argv.length < 3) {
     console.error(`Specify an image file path as the first argument and a second image file path as the second (optional) argument`);
     process.exit(1);
@@ -8,11 +8,11 @@ try {
     const imageFilePath = process.argv[2];
     // Load the main model with projection in a single step
     const modelId = await loadModel({
-        modelSrc: SMOLVLM2_500M_MULTIMODAL_Q8_0,
+        modelSrc: SMOLVLM2_2_500M_MULTIMODAL_Q8_0,
         modelType: "llm",
+        projectionModelSrc: MMPROJ_SMOLVLM2_2_500M_MULTIMODAL_Q8_0,
         modelConfig: {
             ctx_size: 1024,
-            projectionModelSrc: MMPROJ_SMOLVLM2_500M_MULTIMODAL_Q8_0,
         },
         onProgress: (progress) => {
             console.log(`Loading: ${progress.percentage.toFixed(1)}%`);
@@ -54,6 +54,7 @@ try {
     console.log("\n📊 Performance Stats:", stats2);
     console.log("--------------------------------");
     await unloadModel({ modelId, clearStorage: false });
+    process.exit(0);
 }
 catch (error) {
     console.error("❌ Error:", error);