@qvac/sdk 0.12.1 → 0.13.0

package/dist/examples/bci/bci-filesystem-streaming.js

@@ -0,0 +1,77 @@
+/**
+ * Streaming BCI transcription from a neural-signal file.
+ *
+ * Reads a raw neural-signal `.bin` file and feeds it to the BCI
+ * (whisper.cpp) addon chunk-by-chunk through a duplex `bciTranscribeStream`
+ * session, printing transcript text as the sliding window decodes
+ * successive windows.
+ *
+ * Usage: bun run examples/bci/bci-filesystem-streaming.ts <neural-bin-file-path>
+ */
+import { loadModel, unloadModel, bciTranscribeStream, BCI_WINDOWED, } from "@qvac/sdk";
+import { readFileSync } from "fs";
+const args = process.argv.slice(2);
+if (!args[0]) {
+    console.error("Usage: bun run examples/bci/bci-filesystem-streaming.ts <neural-bin-file-path>");
+    process.exit(1);
+}
+const neuralFilePath = args[0];
+// Feed the neural buffer in fixed-size chunks to simulate a live stream.
+const CHUNK_SIZE = 64 * 1024;
+try {
+    console.log("=== BCI transcribeStream file test ===");
+    console.log(`File: ${neuralFilePath}`);
+    console.log(`Chunk size: ${CHUNK_SIZE} bytes\n`);
+    console.log("Loading model...");
+    const modelId = await loadModel({
+        modelSrc: BCI_WINDOWED,
+        modelConfig: {
+            whisperConfig: {
+                language: "en",
+                n_threads: 4,
+                temperature: 0.0,
+            },
+            // Session day index selects the day-specific projection matrices.
+            // Set this to match the recording session your neural file came from.
+            bciConfig: {
+                day_idx: 1,
+            },
+        },
+    });
+    console.log(`Model loaded: ${modelId}\n`);
+    console.log("Opening live session...");
+    const session = await bciTranscribeStream({ modelId, emit: "delta" });
+    console.log("Session open. Streaming neural signal...\n");
+    // Drain the session concurrently with writing so the sliding-window
+    // decode can make progress as chunks arrive instead of stalling.
+    const consume = (async () => {
+        let transcript = "";
+        for await (const text of session) {
+            transcript += text;
+            process.stdout.write(text);
+        }
+        return transcript;
+    })();
+    const data = readFileSync(neuralFilePath);
+    let totalBytes = 0;
+    for (let offset = 0; offset < data.length; offset += CHUNK_SIZE) {
+        const chunk = data.subarray(offset, offset + CHUNK_SIZE);
+        session.write(chunk);
+        totalBytes += chunk.length;
+        await new Promise((resolve) => setTimeout(resolve, 10));
+    }
+    console.log(`\n\nNeural signal streamed: ${totalBytes} bytes`);
+    console.log("Waiting for transcription to finish...\n");
+    session.end();
+    const transcript = await consume;
+    console.log("\n=== Results ===");
+    console.log(`Transcript: ${transcript.trim() || "(no text received)"}`);
+    console.log("\nUnloading model...");
+    await unloadModel({ modelId });
+    console.log("Done.");
+    process.exit(0);
+}
+catch (error) {
+    console.error("❌ Error:", error);
+    process.exit(1);
+}

package/dist/examples/bci/bci-filesystem.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=bci-filesystem.d.ts.map

package/dist/examples/bci/bci-filesystem.js

@@ -0,0 +1,63 @@
+/**
+ * Batch BCI transcription from a neural-signal file.
+ *
+ * Reads a raw neural-signal `.bin` file, runs it through the BCI
+ * (whisper.cpp) addon in one shot via `bciTranscribe`, and prints the
+ * decoded transcript.
+ *
+ * Usage: bun run examples/bci/bci-filesystem.ts <neural-bin-file-path>
+ */
+import { loadModel, unloadModel, bciTranscribe, BCI_WINDOWED } from "@qvac/sdk";
+const args = process.argv.slice(2);
+if (!args[0]) {
+    console.error("Usage: bun run examples/bci/bci-filesystem.ts <neural-bin-file-path>");
+    process.exit(1);
+}
+const neuralFilePath = args[0];
+try {
+    console.log("🧠 Starting BCI transcription example...");
+    console.log("📥 Loading BCI model...");
+    const modelId = await loadModel({
+        modelSrc: BCI_WINDOWED,
+        modelConfig: {
+            whisperConfig: {
+                language: "en",
+                n_threads: 4,
+                temperature: 0.0,
+            },
+            // Session day index selects the day-specific projection matrices.
+            // Set this to match the recording session your neural file came from.
+            bciConfig: {
+                day_idx: 1,
+            },
+        },
+        onProgress: (progress) => {
+            console.log(progress);
+        },
+    });
+    console.log(`✅ BCI model loaded with ID: ${modelId}`);
+    console.log("🧠 Transcribing neural signal...");
+    const segments = await bciTranscribe({
+        modelId,
+        neuralData: neuralFilePath,
+        metadata: true,
+    });
+    console.log("📝 Transcription result:");
+    for (const segment of segments) {
+        const start = (segment.startMs / 1000).toFixed(2);
+        const end = (segment.endMs / 1000).toFixed(2);
+        console.log(`  [${start}s → ${end}s] (id=${segment.id}, append=${segment.append}) ${segment.text}`);
+    }
+    console.log(`\nFull transcript: ${segments
+        .map((s) => s.text)
+        .join("")
+        .trim()}`);
+    console.log("🧹 Unloading BCI model...");
+    await unloadModel({ modelId });
+    console.log("✅ BCI model unloaded successfully");
+    process.exit(0);
+}
+catch (error) {
+    console.error("❌ Error:", error);
+    process.exit(1);
+}

package/dist/examples/cache-management.js

@@ -56,7 +56,6 @@ try {
         console.log("   Loading model into memory...");
         modelId = await loadModel({
             modelSrc: WHISPER_TINY,
-            modelType: "whisper",
         });
         console.log(`   ✅ Model loaded! ID: ${modelId}`);
         const afterLoad = await getModelInfo(WHISPER_TINY);

package/dist/examples/cancel-by-request-id.js

@@ -38,7 +38,6 @@ import { cancel, completion, loadModel, unloadModel, InferenceCancelledError, QW
 try {
     const modelId = await loadModel({
         modelSrc: QWEN3_600M_INST_Q4,
-        modelType: "llm",
         modelConfig: { ctx_size: 4096 },
     });
     const run = completion({

package/dist/examples/classification/classify-image.js

@@ -1,5 +1,5 @@
 import fs from "fs";
-import { startQVACProvider, stopQVACProvider, loadModel, classify, unloadModel } from "@qvac/sdk";
+import { startQVACProvider, stopQVACProvider, loadModel, classify, unloadModel, } from "@qvac/sdk";
 /**
  * Classify an image using the bundled MobileNetV3-Small model.
  *

package/dist/examples/completion-events.js

@@ -18,14 +18,15 @@ import { completion, loadModel, unloadModel, QWEN3_600M_INST_Q4, } from "@qvac/s
 try {
     const modelId = await loadModel({
         modelSrc: QWEN3_600M_INST_Q4,
-        modelType: "llm",
         modelConfig: { ctx_size: 4096 },
         onProgress: (p) => console.log(`Loading: ${p.percentage.toFixed(1)}%`),
     });
     console.log(`✅ Model loaded: ${modelId}\n`);
     const result = completion({
         modelId,
-        history: [{ role: "user", content: "Explain quantum computing in 2 sentences" }],
+        history: [
+            { role: "user", content: "Explain quantum computing in 2 sentences" },
+        ],
         stream: true,
         captureThinking: true,
     });
@@ -44,6 +45,9 @@ try {
     if (final.toolCalls.length > 0) {
         console.log(`Tool calls: ${final.toolCalls.map((c) => c.name).join(", ")}`);
     }
+    if (final.stopReason) {
+        console.log(`Stop reason: ${final.stopReason}`);
+    }
     console.log(`Raw output length: ${final.raw.fullText.length} chars`);
     await unloadModel({ modelId, clearStorage: false });
 }

package/dist/examples/completion-stop-reason.d.ts

@@ -0,0 +1,25 @@
+/**
+ * Demonstrates `stopReason` on `CompletionFinal`.
+ *
+ * `completion()` sets `stopReason` on the aggregated `final` promise to
+ * indicate why the model stopped generating:
+ *
+ *  - `undefined`    — natural end-of-sequence (EOS). The model finished on
+ *                     its own. This is the common case and the backwards-
+ *                     compatible default.
+ *  - `"length"`     — the `predict` token budget was exhausted. Output is
+ *                     truncated; the model did not reach a natural stopping
+ *                     point.
+ *  - `"cancelled"`  — the request was cancelled via `cancel({ requestId })`.
+ *                     See `cancel-by-request-id.ts` for the full cancel flow.
+ *
+ * This example shows the two non-cancel cases back-to-back on the same model:
+ *
+ *  1. Natural EOS  — no `generationParams`, model stops on its own.
+ *  2. Budget hit   — `predict: 10` forces truncation mid-output.
+ *
+ * The same `stopReason` is also available on the terminal `completionDone`
+ * event in the `events` stream.
+ */
+export {};
+//# sourceMappingURL=completion-stop-reason.d.ts.map

package/dist/examples/completion-stop-reason.js

@@ -0,0 +1,65 @@
+/**
+ * Demonstrates `stopReason` on `CompletionFinal`.
+ *
+ * `completion()` sets `stopReason` on the aggregated `final` promise to
+ * indicate why the model stopped generating:
+ *
+ *  - `undefined`    — natural end-of-sequence (EOS). The model finished on
+ *                     its own. This is the common case and the backwards-
+ *                     compatible default.
+ *  - `"length"`     — the `predict` token budget was exhausted. Output is
+ *                     truncated; the model did not reach a natural stopping
+ *                     point.
+ *  - `"cancelled"`  — the request was cancelled via `cancel({ requestId })`.
+ *                     See `cancel-by-request-id.ts` for the full cancel flow.
+ *
+ * This example shows the two non-cancel cases back-to-back on the same model:
+ *
+ *  1. Natural EOS  — no `generationParams`, model stops on its own.
+ *  2. Budget hit   — `predict: 10` forces truncation mid-output.
+ *
+ * The same `stopReason` is also available on the terminal `completionDone`
+ * event in the `events` stream.
+ */
+import { completion, loadModel, unloadModel, QWEN3_600M_INST_Q4, } from "@qvac/sdk";
+try {
+    const modelId = await loadModel({
+        modelSrc: QWEN3_600M_INST_Q4,
+        modelConfig: { ctx_size: 4096 },
+        onProgress: (p) => console.log(`Loading: ${p.percentage.toFixed(1)}%`),
+    });
+    console.log(`✅ Model loaded: ${modelId}\n`);
+    // --- Case 1: natural EOS ---
+    console.log("=== Case 1: natural EOS (no predict limit) ===");
+    const run1 = completion({
+        modelId,
+        history: [{ role: "user", content: "Say hi in one word." }],
+        stream: true,
+    });
+    for await (const event of run1.events) {
+        if (event.type === "contentDelta")
+            process.stdout.write(event.text);
+    }
+    const final1 = await run1.final;
+    console.log(`\nstopReason: ${final1.stopReason ?? "(undefined — natural EOS)"}`);
+    // --- Case 2: predict budget exhausted → stopReason = "length" ---
+    console.log("\n=== Case 2: predict budget hit (predict: 10) ===");
+    const run2 = completion({
+        modelId,
+        history: [{ role: "user", content: "Count from 1 to 100, one number per line." }],
+        generationParams: { predict: 10 },
+        stream: true,
+    });
+    for await (const event of run2.events) {
+        if (event.type === "contentDelta")
+            process.stdout.write(event.text);
+    }
+    const final2 = await run2.final;
+    console.log(`\nstopReason: ${final2.stopReason}`);
+    await unloadModel({ modelId, clearStorage: false });
+    process.exit(0);
+}
+catch (error) {
+    console.error("❌ Error:", error);
+    process.exit(1);
+}

package/dist/examples/concurrent-completion-collision.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=concurrent-completion-collision.d.ts.map

package/dist/examples/concurrent-completion-collision.js

@@ -0,0 +1,98 @@
+/**
+ * Demonstrates how `qvac serve` handles a coding agent firing two completions
+ * at the SAME loaded model at once (e.g. a chat completion plus a
+ * title/summary call) — the workload that used to break it.
+ *
+ * Background: a loaded llama.cpp model is one native context (one KV-cache,
+ * one decode loop), so two completions on it cannot truly run in parallel.
+ * The SDK worker installs a per-`(kind, modelId)` FIFO admission queue for
+ * the `completion` kind (`maxConcurrentPerModel: 1`, `onOverflow: "queue"`).
+ * A second concurrent completion on the same model now WAITS its turn and
+ * then runs, instead of being rejected with `RequestRejectedByPolicyError`
+ * (code 52420) or colliding at the addon with "Cannot set new job".
+ *
+ * Part 1 — two concurrent completions on the SAME model  → both succeed,
+ *          serialized by the queue (the bug this fixes: previously one was
+ *          rejected/collided).
+ * Part 2 — two concurrent completions on DIFFERENT models → both succeed and
+ *          run in parallel (the queue is keyed per `(kind, modelId)`, so
+ *          distinct models never block each other).
+ *
+ * Run from packages/sdk:
+ *   bun run examples/concurrent-completion-collision.ts
+ */
+import { completion, loadModel, unloadModel, QWEN3_1_7B_INST_Q4, LLAMA_3_2_1B_INST_Q4_0, RequestRejectedByPolicyError, } from "@qvac/sdk";
+async function runCompletion(label, modelId, prompt) {
+    try {
+        // `completion()` starts the request synchronously, so calling this for two
+        // tasks in the same tick puts both "in flight" before either is awaited.
+        const result = completion({
+            modelId,
+            history: [{ role: "user", content: prompt }],
+            stream: false,
+        });
+        const text = await result.text;
+        return { label, ok: true, text };
+    }
+    catch (error) {
+        const code = typeof error === "object" && error !== null && "code" in error
+            ? error.code
+            : undefined;
+        const base = {
+            label,
+            ok: false,
+            errorName: error instanceof Error ? error.constructor.name : typeof error,
+            message: error instanceof Error ? error.message : String(error),
+        };
+        return code !== undefined ? { ...base, code } : base;
+    }
+}
+function report(title, outcomes) {
+    console.log(`\n${title}`);
+    for (const o of outcomes) {
+        if (o.ok) {
+            const preview = o.text.replace(/\s+/g, " ").slice(0, 70);
+            console.log(`  ✅ ${o.label}: ${preview}${o.text.length > 70 ? "…" : ""}`);
+        }
+        else {
+            const policy = o.code === 52420 ? " (rejected by policy)" : "";
+            console.log(`  ❌ ${o.label}: ${o.errorName}${o.code !== undefined ? ` [${o.code}]` : ""}${policy} — ${o.message}`);
+        }
+    }
+}
+try {
+    const modelA = await loadModel({
+        modelSrc: QWEN3_1_7B_INST_Q4,
+        modelConfig: { ctx_size: 4096 },
+    });
+    // Part 1 — same model, two concurrent completions.
+    const sameModel = await Promise.all([
+        runCompletion("req-A1", modelA, "What is Bitcoin? One sentence."),
+        runCompletion("req-A2", modelA, "What is a hash function? One sentence."),
+    ]);
+    report("Part 1 — two concurrent completions on the SAME model (expect both to succeed, serialized):", sameModel);
+    const rejected = sameModel.filter((o) => !o.ok).length;
+    if (rejected === 0) {
+        console.log("  → Both succeeded: the second was queued behind the first and ran when the slot freed (no 52420, no addon collision).");
+    }
+    else {
+        console.log(`  ⚠️  ${rejected}/2 failed — the FIFO admission queue should have serialized these. Check the completion policy.`);
+    }
+    // Part 2 — different models, two concurrent completions (the "parallel that works").
+    const modelB = await loadModel({
+        modelSrc: LLAMA_3_2_1B_INST_Q4_0,
+        modelConfig: { ctx_size: 4096 },
+    });
+    const differentModels = await Promise.all([
+        runCompletion("req-A", modelA, "Name one use of Bitcoin. One sentence."),
+        runCompletion("req-B", modelB, "Name one use of Ethereum. One sentence."),
+    ]);
+    report("Part 2 — two concurrent completions on DIFFERENT models (expect both succeed):", differentModels);
+    await unloadModel({ modelId: modelA, clearStorage: false });
+    await unloadModel({ modelId: modelB, clearStorage: false });
+    console.log(`\nrejected-by-policy error available for instanceof checks: ${typeof RequestRejectedByPolicyError === "function"}`);
+}
+catch (error) {
+    console.error("❌ Error:", error);
+    process.exit(1);
+}

package/dist/examples/config-reload.js

@@ -11,7 +11,6 @@ try {
     console.log("Loading Whisper model with English config...");
     const modelId = await loadModel({
         modelSrc: WHISPER_TINY,
-        modelType: "whisper",
         modelConfig: {
             language: "en",
         },
@@ -28,7 +27,7 @@ try {
     console.log("Hot reloading config (changing language and temperature)...");
     const reloadedId = await loadModel({
         modelId,
-        modelType: "whisper",
+        modelType: "whispercpp-transcription",
         modelConfig: {
             language: "es", // Change to Spanish
         },

package/dist/examples/default-config-usage.js

@@ -24,7 +24,6 @@ try {
     console.log("🚀 Loading model with configured settings...\n");
     const modelId = await loadModel({
         modelSrc: LLAMA_3_2_1B_INST_Q4_0,
-        modelType: "llm",
     });
     console.log("\n💬 Running completion...\n");
     const result = completion({

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

@@ -56,7 +56,6 @@ async function runDelegatedCompletion(providerPublicKey) {
     console.log("→ Loading model via delegation...");
     const modelId = await loadModel({
         modelSrc: LLAMA_3_2_1B_INST_Q4_0,
-        modelType: "llm",
         delegate: {
             providerPublicKey,
             timeout: 30_000,

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

@@ -11,7 +11,6 @@ try {
     console.log("→ Loading model (delegated, unary)...");
     const modelId = await loadModel({
         modelSrc: LLAMA_3_2_1B_INST_Q4_0,
-        modelType: "llm",
         delegate: {
             providerPublicKey,
             timeout: 60_000,

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

@@ -18,7 +18,6 @@ try {
     }
     const modelId = await loadModel({
         modelSrc: LLAMA_3_2_1B_INST_Q4_0,
-        modelType: "llm",
         delegate: {
             providerPublicKey,
             // Generous timeout for the first call on a cold DHT: bootstrapping

package/dist/examples/diffusion-esrgan-upscale.js

@@ -8,8 +8,7 @@ import path from "path";
 const esrganArg = process.argv[2];
 const promptArg = process.argv[3];
 const outputDirArg = process.argv[4];
-const esrganModelSrc = esrganArg ??
-    REALESRGAN_X4PLUS_ANIME_6B;
+const esrganModelSrc = esrganArg ?? REALESRGAN_X4PLUS_ANIME_6B;
 const prompt = promptArg ??
     "an illustrated red fox portrait, clean line art, soft watercolor background, detailed fur, crisp eyes";
 const negative_prompt = "blurry, low quality, watermark, text";
@@ -19,7 +18,6 @@ try {
     console.log("Loading SD 2.1 + ESRGAN upscaler...");
     const modelId = await loadModel({
         modelSrc: SD_V2_1_1B_Q8_0,
-        modelType: "diffusion",
         modelConfig: {
             prediction: "v",
             upscaler: {

package/dist/examples/diffusion-flux2-klein-img2img.js

@@ -17,7 +17,7 @@ try {
     console.log("Loading FLUX.2 [klein] split-layout model...");
     const modelId = await loadModel({
         modelSrc: diffusionModelSrc,
-        modelType: "diffusion",
+        modelType: "sdcpp-generation",
         modelConfig: {
             device: "gpu",
             threads: 4,

package/dist/examples/diffusion-flux2-klein.js

@@ -1,4 +1,4 @@
-import { loadModel, unloadModel, diffusion, FLUX_2_KLEIN_4B_Q4_0, FLUX_2_KLEIN_4B_VAE, QWEN3_4B_Q4_K_M } from "@qvac/sdk";
+import { loadModel, unloadModel, diffusion, FLUX_2_KLEIN_4B_Q4_0, FLUX_2_KLEIN_4B_VAE, QWEN3_4B_Q4_K_M, } from "@qvac/sdk";
 import fs from "fs";
 import path from "path";
 // FLUX.2 [klein] uses a split-layout: separate diffusion model + LLM text encoder + VAE
@@ -10,7 +10,7 @@ const outputDir = process.argv[6] || ".";
 console.log("Loading FLUX.2 [klein] split-layout model...");
 const modelId = await loadModel({
     modelSrc: diffusionModelSrc,
-    modelType: "diffusion",
+    modelType: "sdcpp-generation",
     modelConfig: {
         device: "gpu",
         threads: 4,

package/dist/examples/diffusion-img2img.js

@@ -15,7 +15,7 @@ try {
     console.log("Loading diffusion model...");
     const modelId = await loadModel({
         modelSrc,
-        modelType: "diffusion",
+        modelType: "sdcpp-generation",
     });
     console.log(`Model loaded: ${modelId}`);
     const init_image = new Uint8Array(fs.readFileSync(inputPath));

package/dist/examples/diffusion-img2vid.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=diffusion-img2vid.d.ts.map

package/dist/examples/diffusion-img2vid.js

@@ -0,0 +1,78 @@
+import { loadModel, unloadModel, video, WAN2_1_I2V_14B_Q4_K_M, CLIP_VISION_H, UMT5_XXL_FP16, WAN_2_1_COMFYUI_REPACKAGED_VAE, } from "@qvac/sdk";
+import fs from "fs";
+import path from "path";
+// Image-to-video with Wan 2.1 I2V. Requires a Wan I2V diffusion checkpoint (GGUF
+// recommended), plus UMT5-XXL, Wan VAE, and CLIP vision weights. The model
+// sources default to the bundled registry constants, so the common case is just
+// an init image path.
+const initImagePath = process.argv[2];
+const prompt = process.argv[3] ||
+    "the subject slowly turns and smiles, soft natural lighting, cinematic";
+const outputDir = process.argv[4] || ".";
+const diffusionModelSrc = process.argv[5] || WAN2_1_I2V_14B_Q4_K_M;
+const t5XxlModelSrc = process.argv[6] || UMT5_XXL_FP16;
+const vaeModelSrc = process.argv[7] || WAN_2_1_COMFYUI_REPACKAGED_VAE;
+const clipVisionModelSrc = process.argv[8] || CLIP_VISION_H;
+if (!initImagePath) {
+    console.error("❌ Error: init image path is required");
+    console.error("Usage: bun run bare:example dist/examples/diffusion-img2vid.js " +
+        "<initImagePath> [prompt] [outputDir] " +
+        "[i2vModelSrc] [t5XxlModelSrc] [vaeModelSrc] [clipVisionModelSrc]");
+    process.exit(1);
+}
+try {
+    console.log("Loading Wan 2.1 I2V model (diffusion + UMT5-XXL + VAE + CLIP vision)...");
+    const modelId = await loadModel({
+        modelSrc: diffusionModelSrc,
+        modelType: "diffusion",
+        modelConfig: {
+            mode: "video",
+            device: "gpu",
+            threads: 4,
+            t5XxlModelSrc,
+            vaeModelSrc,
+            clipVisionModelSrc,
+            diffusion_fa: true,
+            offload_to_cpu: true,
+            vae_on_cpu: true,
+            vae_tiling: true,
+        },
+        onProgress: (p) => console.log(`Loading: ${p.percentage.toFixed(1)}%`),
+    });
+    console.log(`Model loaded: ${modelId}`);
+    const init_image = new Uint8Array(fs.readFileSync(initImagePath));
+    console.log(`\nGenerating video for: "${prompt}"`);
+    const { progressStream, outputs, stats } = video({
+        modelId,
+        mode: "img2vid",
+        prompt,
+        init_image,
+        negative_prompt: "blurry, distorted, low quality, jittery, static, frozen",
+        strength: 0.85,
+        flow_shift: 3.0,
+        video_frames: 33,
+        fps: 16,
+        steps: 30,
+        cfg_scale: 6.0,
+        seed: 42,
+        vae_tiling: true,
+    });
+    for await (const { step, totalSteps } of progressStream) {
+        process.stdout.write(`\rStep ${step}/${totalSteps}`);
+    }
+    console.log();
+    const buffers = await outputs;
+    for (let i = 0; i < buffers.length; i++) {
+        const outputPath = path.join(outputDir, `wan_i2v_${i}.avi`);
+        fs.writeFileSync(outputPath, buffers[i]);
+        console.log(`Saved: ${outputPath}`);
+    }
+    console.log("\nStats:", await stats);
+    await unloadModel({ modelId, clearStorage: false });
+    console.log("Done.");
+    process.exit(0);
+}
+catch (error) {
+    console.error("❌ Error:", error);
+    process.exit(1);
+}

package/dist/examples/diffusion-simple.js

@@ -6,7 +6,7 @@ const modelSrc = process.argv[2] || SD_V2_1_1B_Q8_0;
 const prompt = process.argv[3] || "a photo of a cat sitting on a windowsill";
 const modelId = await loadModel({
     modelSrc,
-    modelType: "diffusion",
+    modelType: "sdcpp-generation",
     modelConfig: { prediction: "v" },
 });
 const { outputs } = diffusion({ modelId, prompt });

package/dist/examples/diffusion-txt2img.js

@@ -1,4 +1,4 @@
-import { loadModel, unloadModel, diffusion, FLUX_2_KLEIN_4B_Q4_0, FLUX_2_KLEIN_4B_VAE, QWEN3_4B_Q4_K_M } from "@qvac/sdk";
+import { loadModel, unloadModel, diffusion, FLUX_2_KLEIN_4B_Q4_0, FLUX_2_KLEIN_4B_VAE, QWEN3_4B_Q4_K_M, } from "@qvac/sdk";
 import fs from "fs";
 import path from "path";
 const modelSrc = process.argv[2] || FLUX_2_KLEIN_4B_Q4_0;
@@ -9,8 +9,13 @@ console.log(`Loading diffusion model...`);
 // FLUX.2 models require companion LLM + VAE models
 const modelId = await loadModel({
     modelSrc,
-    modelType: "diffusion",
-    modelConfig: { device: "gpu", threads: 4, llmModelSrc: QWEN3_4B_Q4_K_M, vaeModelSrc: FLUX_2_KLEIN_4B_VAE },
+    modelType: "sdcpp-generation",
+    modelConfig: {
+        device: "gpu",
+        threads: 4,
+        llmModelSrc: QWEN3_4B_Q4_K_M,
+        vaeModelSrc: FLUX_2_KLEIN_4B_VAE,
+    },
     onProgress: (p) => console.log(`Loading: ${p.percentage.toFixed(1)}%`),
 });
 console.log(`Model loaded: ${modelId}`);

package/dist/examples/diffusion-txt2vid.js

@@ -11,14 +11,13 @@ const vaeModelSrc = process.argv[4] || WAN_2_1_COMFYUI_REPACKAGED_VAE;
 // Prompt tip: Wan 1.3B is small and has weak temporal priors. Use motion-
 // explicit verbs and avoid static framing words like "standing", "still",
 // or "portrait" in the positive prompt.
-const prompt = process.argv[5] ||
-    "a colorful bird flapping its wings";
+const prompt = process.argv[5] || "a colorful bird flapping its wings";
 const outputDir = process.argv[6] || ".";
 try {
     console.log("Loading Wan 2.1 T2V model (diffusion + UMT5-XXL + VAE)...");
     const modelId = await loadModel({
         modelSrc: diffusionModelSrc,
-        modelType: "diffusion",
+        modelType: "sdcpp-generation",
         modelConfig: {
             mode: "video",
             device: "gpu",

package/dist/examples/download-with-blind-relays.js

@@ -15,7 +15,6 @@ try {
     const startTime = Date.now();
     const modelId = await loadModel({
         modelSrc: LLAMA_3_2_1B_INST_Q4_0,
-        modelType: "llm",
     });
     console.log(`Model loaded with ID: ${modelId}`);
     const firstStatus = await getModelInfo(LLAMA_3_2_1B_INST_Q4_0);

package/dist/examples/embed-multi-gpu.js

@@ -1,4 +1,4 @@
-import { loadModel, embed, unloadModel, EMBEDDINGGEMMA_300M_Q8_0 } from "@qvac/sdk";
+import { loadModel, embed, unloadModel, EMBEDDINGGEMMA_300M_Q8_0, } from "@qvac/sdk";
 // Multi-GPU embedding distributes model layers across multiple GPUs using
 // llama.cpp's built-in split modes. Two strategies are available:
 //
@@ -15,7 +15,6 @@ import { loadModel, embed, unloadModel, EMBEDDINGGEMMA_300M_Q8_0 } from "@qvac/s
 try {
     const modelId = await loadModel({
         modelSrc: EMBEDDINGGEMMA_300M_Q8_0,
-        modelType: "llamacpp-embedding",
         modelConfig: {
             splitMode: "layer",
             tensorSplit: "1,1",