Migrate workspace to Vite+ 1.0 preview

.gitignore

@@ -218,3 +218,4 @@ __pycache__/
 *.metallib
 .vite-agent
 /Model-Optimizer
+.vitest/

.yarnrc.yml

@@ -1,11 +1,10 @@
 approvedGitRepositories:
-  - "**"
+  - '**'
 
 catalog:
-  "@types/node": "npm:@types/node@26.4.0"
-  vite: "npm:@voidzero-dev/vite-plus-core@0.3.0"
-  vite-plus: 0.3.0
-  vitest: "npm:vitest@4.1.11"
+  '@types/node': 'npm:@types/node@26.4.0'
+  vite: npm:@voidzero-dev/vite-plus-core@0.0.0-commit.e0f77ee2353b38c77d7165d3305f9f6d705b4d62
+  vite-plus: 0.0.0-commit.e0f77ee2353b38c77d7165d3305f9f6d705b4d62
 
 enableScripts: true
 
@@ -15,6 +14,10 @@ npmMinimalAgeGate: 0
 
 npmScopes:
   jsr:
-    npmRegistryServer: "https://npm.jsr.io"
+    npmRegistryServer: 'https://npm.jsr.io'
 
 yarnPath: .yarn/releases/yarn-4.18.0.cjs
+npmPreapprovedPackages:
+  - vitest
+  - '@vitest/*'
+npmRegistryServer: https://registry-bridge.viteplus.dev/

Cargo.toml

@@ -23,11 +23,7 @@ serde = { version = "1.0.229", features = ["derive"] }
 thiserror = "2.0.20"
 tokio = { version = "1.53", features = ["rt-multi-thread", "macros", "sync"] }
 tracing = "0.1.44"
-tracing-subscriber = { version = "0.3.23", default-features = false, features = [
-  "std",
-  "fmt",
-  "json",
-] }
+tracing-subscriber = { version = "0.3.23", default-features = false, features = ["std", "fmt", "json"] }
 tokenizers = "0.23.1"
 minijinja = { version = "2.24", features = ["json", "preserve_order"] }
 uuid = { version = "1.26", features = ["v4"] }

__test__/cli/serve.test.ts

@@ -1,4 +1,5 @@
 import { mkdir, mkdtemp, rm, writeFile } from 'node:fs/promises';
+import { connect } from 'node:net';
 import { tmpdir } from 'node:os';
 import { join } from 'node:path';
 
@@ -164,7 +165,15 @@ describe('mlx serve — end to end', () => {
 
     expect(observedUrl).toMatch(/^http:\/\/127\.0\.0\.1:\d+$/);
     expect(exitCodes).toEqual([0]);
-    await expect(fetch(healthUrl)).rejects.toThrow();
+    const connectionError = await new Promise<NodeJS.ErrnoException | null>((resolve) => {
+      const socket = connect(Number(new URL(healthUrl).port), '127.0.0.1');
+      socket.once('error', (error) => resolve(error as NodeJS.ErrnoException));
+      socket.once('connect', () => {
+        socket.destroy();
+        resolve(null);
+      });
+    });
+    expect(connectionError?.code).toBe('ECONNREFUSED');
 
     // Handlers detached on shutdown: a leaked one would keep a closed host
     // alive in the listener list and swallow the next Ctrl+C.

__test__/models/chat-session-mtp-default.test.ts

@@ -45,8 +45,8 @@
 
 /** The session surface every case needs, minus the MTP getters. */
 function sessionStubs() {
-  const chatSessionStart = vi.fn(
-    async (_messages: ChatMessage[], _config?: ChatConfig | null): Promise<ChatResult> => makeChatResult('reply'),
+  const chatSessionStart = vi.fn(async (_messages: ChatMessage[], _config?: ChatConfig | null): Promise<ChatResult> =>
+    makeChatResult('reply'),
   );
   const chatSessionContinue = vi.fn(
     async (_messages: ChatMessage[], _config?: ChatConfig | null): Promise<ChatResult> => makeChatResult('reply'),
@@ -231,6 +231,6 @@
     // Collapsing the two into one would make the head unreachable rather than
     // merely off-by-default.
     expect(typeof NemotronHModel.prototype.hasMtpWeights).toBe('function');
     expect(NemotronHModel.prototype.mtpAutoEnabled).not.toBe(NemotronHModel.prototype.hasMtpWeights);
   });
 });

__test__/models/gemma4-concurrency-e2e.test.ts

@@ -27,9 +27,9 @@ function isScheduledGemma(model: LoadableModel): model is LoadableModel & Schedu
   );
 }
 
-const gemmaDescribe = MODEL_ENV_PRESENT ? describe.sequential : describe.skip;
+const gemmaDescribe = MODEL_ENV_PRESENT ? describe : describe.skip;
 
-gemmaDescribe('Gemma4 grouped hybrid KV continuous batching', () => {
+gemmaDescribe('Gemma4 grouped hybrid KV continuous batching', { concurrent: false }, () => {
   let model: ScheduledGemma;
   const sessions: ChatSession[] = [];
   const activeSends = new Set<{

__test__/models/gemma4-unified-detection.test.ts

@@ -13,7 +13,7 @@ import { afterEach, beforeEach, describe, expect, it } from 'vite-plus/test';
  * resolve to the one shared `gemma4` Rust decoder + `Gemma4Model` registry
  * row — there is no separate unified registry entry.
  */
-describe.sequential('Gemma4 unified model detection', () => {
+describe('Gemma4 unified model detection', { concurrent: false }, () => {
   let tempDir: string;
 
   beforeEach(async () => {

__test__/models/harrier-detection.test.ts

@@ -5,7 +5,7 @@ import { join } from 'node:path';
 import { detectModelType } from '@mlx-node/lm';
 import { describe, it, expect, beforeEach, afterEach } from 'vite-plus/test';
 
-describe.sequential('Harrier Model Detection', () => {
+describe('Harrier Model Detection', { concurrent: false }, () => {
   let tempDir: string;
 
   beforeEach(async () => {

__test__/models/harrier.test.ts

@@ -33,7 +33,7 @@ const HARRIER_0_6B_CONFIG = {
   useQkNorm: true,
 };
 
-describe.sequential('HarrierModel', () => {
+describe('HarrierModel', { concurrent: false }, () => {
   describe('Model Instantiation', () => {
     it('should create model from tiny config', () => {
       const model = new HarrierModel(TINY_CONFIG);

__test__/models/lfm2-moe-detection.test.ts

@@ -5,7 +5,7 @@ import { join } from 'node:path';
 import { detectModelType } from '@mlx-node/lm';
 import { describe, it, expect, beforeEach, afterEach } from 'vite-plus/test';
 
-describe.sequential('LFM2 MoE Model Detection', () => {
+describe('LFM2 MoE Model Detection', { concurrent: false }, () => {
   let tempDir: string;
 
   beforeEach(async () => {

__test__/models/model-loader-registry.test.ts

@@ -42,7 +42,7 @@ const UNIFIED_GEMMA_COMPATIBILITY_CASES = [
   ['competing Harrier probe', { model_type: 'qwen3', architectures: ['Qwen3Model', UNIFIED_GEMMA_ARCHITECTURE] }],
 ] as const satisfies readonly (readonly [string, Readonly<Record<string, unknown>>])[];
 
-describe.sequential('declarative model loader registry', () => {
+describe('declarative model loader registry', { concurrent: false }, () => {
   let tempDir: string;
 
   beforeEach(async () => {

__test__/models/nemotron-h-detection.test.ts

@@ -12,7 +12,7 @@ import { afterEach, beforeEach, describe, expect, it } from 'vite-plus/test';
  * (mirroring the native loader's config validation in
  * crates/mlx-core/src/models/nemotron_h/config.rs).
  */
-describe.sequential('Nemotron H model detection', () => {
+describe('Nemotron H model detection', { concurrent: false }, () => {
   let tempDir: string;
 
   beforeEach(async () => {

__test__/models/qwen3-chat.test.ts

@@ -19,7 +19,7 @@ import { describe, it, expect, beforeAll, afterAll } from 'vite-plus/test';
 
 import { createTempModel, findTokenizerPath, TINY_TEST_CONFIG } from '../test-model-utils';
 
-describe.sequential('Qwen3 Chat Session API', () => {
+describe('Qwen3 Chat Session API', { concurrent: false }, () => {
   let model: Qwen3Model;
   let cleanup: () => void;
 

__test__/models/qwen3-generate-batch-parallel.test.ts

@@ -13,7 +13,7 @@ import { describe, it, expect, beforeAll } from 'vite-plus/test';
  * Each test creates its own GrpoTrainingEngine and resets it when done
  * to release the model thread's training state for the next test.
  */
-describe.sequential('GRPOTrainer - Parallel Batch Generation', () => {
+describe('GRPOTrainer - Parallel Batch Generation', { concurrent: false }, () => {
   let model: Qwen3Model | null = null;
 
   beforeAll(async () => {

__test__/models/qwen3.test.ts

@@ -2,7 +2,7 @@ import { Qwen3Model } from '@mlx-node/core';
 import { QWEN3_CONFIGS, getQwen3Config } from '@mlx-node/lm';
 import { describe, it, expect } from 'vite-plus/test';
 
-describe.sequential('Qwen3 Model', () => {
+describe('Qwen3 Model', { concurrent: false }, () => {
   describe('Model Configuration', () => {
     it('should have correct default configurations', () => {
       expect(QWEN3_CONFIGS['qwen3-0.6b']).toBeDefined();

__test__/models/qwen35-stream.test.ts

@@ -43,7 +43,7 @@ function fakeNativeMethod(numTokens: number) {
   };
 }
 
-describe.sequential('_runChatStream bridge', () => {
+describe('_runChatStream bridge', { concurrent: false }, () => {
   it('should yield delta chunks followed by a final chunk', async () => {
     const events: ChatStreamEvent[] = [];
     const messages: ChatMessage[] = [{ role: 'user', content: 'Hi' }];

__test__/models/qwen3_5-generation.test.ts

@@ -5,7 +5,7 @@ import { afterAll, beforeAll, describe, expect, it } from 'vite-plus/test';
 import { createTempQwen35Model, type TempModel } from '../test-model-utils';
 import { shape } from '../test-utils';
 
-describe.sequential('Qwen3.5 Generation', () => {
+describe('Qwen3.5 Generation', { concurrent: false }, () => {
   let tempModel: TempModel;
 
   beforeAll(async () => {

__test__/server/concurrent-stage0-e2e.test.ts

@@ -330,9 +330,9 @@ function startTickProbe(intervalMs = 50): { stop: () => number } {
   };
 }
 
-const stage0Describe = MODEL_ENV_PRESENT ? describe.sequential : describe.skip;
+const stage0Describe = MODEL_ENV_PRESENT ? describe : describe.skip;
 
-stage0Describe('Stage-0 real-model concurrency hazards', () => {
+stage0Describe('Stage-0 real-model concurrency hazards', { concurrent: false }, () => {
   let instance: ServerInstance;
   let nativeModel: SessionCapableModel;
   let instrumented: InstrumentedModel;

__test__/server/concurrent-stage1-e2e.test.ts

@@ -162,9 +162,9 @@ async function postResponse(baseUrl: string, input: string): Promise<{ id?: stri
   });
 }
 
-const stage1Describe = MODEL_ENV_PRESENT ? describe.sequential : describe.skip;
+const stage1Describe = MODEL_ENV_PRESENT ? describe : describe.skip;
 
-stage1Describe('Stage-1 real-model server admission', () => {
+stage1Describe('Stage-1 real-model server admission', { concurrent: false }, () => {
   let instance: ServerInstance;
   let nativeModel: SchedulerStatsModel;
   let peakActive: () => number;

__test__/server/host/inference-host.test.ts

@@ -1,7 +1,7 @@
 import { spawnSync } from 'node:child_process';
 import { existsSync } from 'node:fs';
 import { chmod, mkdir, mkdtemp, readdir, rm, writeFile } from 'node:fs/promises';
-import { createServer as createNetServer } from 'node:net';
+import { connect as connectNet, createServer as createNetServer } from 'node:net';
 import { tmpdir } from 'node:os';
 import { join } from 'node:path';
 
@@ -324,16 +324,28 @@ describe('createInferenceHost — binding and health', () => {
     const address = host.server.server.address();
     expect(address !== null && typeof address === 'object' ? address.address : null).toBe('::1');
     expect(host.url).toBe(`http://[::1]:${host.port}`);
-    expect((await fetch(`${host.url}/health`)).status).toBe(200);
+    const response = await fetch(`${host.url}/health`);
+    expect(response.status).toBe(200);
+    await response.arrayBuffer();
   });
 
   it('stops answering once closed', async () => {
     const modelsDir = await makeModelsDir(['alpha']);
     const host = await start({ modelsDir });
     const url = `${host.url}/health`;
-    expect((await fetch(url)).status).toBe(200);
+    const response = await fetch(url);
+    expect(response.status).toBe(200);
+    await response.arrayBuffer();
     await host.close({ timeoutMs: 0 });
-    await expect(fetch(url)).rejects.toThrow();
+    const connectionError = await new Promise<NodeJS.ErrnoException | null>((resolve) => {
+      const socket = connectNet(host.port, '127.0.0.1');
+      socket.once('error', (error) => resolve(error as NodeJS.ErrnoException));
+      socket.once('connect', () => {
+        socket.destroy();
+        resolve(null);
+      });
+    });
+    expect(connectionError?.code).toBe('ECONNREFUSED');
   });
 
   /**
@@ -559,7 +571,13 @@ describe('createInferenceHost — a network-reachable bind must be gated', () =>
     const modelsDir = await makeModelsDir(['alpha']);
     for (const bind of ['127.0.0.1', 'localhost', undefined]) {
       const host = await start({ modelsDir, host: bind, port: 0 });
-      expect((await fetch(`${host.url}/v1/models`)).status, String(bind)).toBe(200);
+      expect(host.host).toBe(bind ?? '127.0.0.1');
+      const address = host.server.server.address();
+      if (address === null || typeof address === 'string') throw new Error('Expected an IP socket');
+      const ip = address.family === 'IPv6' ? `[${address.address}]` : address.address;
+      const response = await fetch(`http://${ip}:${address.port}/v1/models`);
+      expect(response.status, String(bind)).toBe(200);
+      await response.arrayBuffer();
     }
   });
 });

__test__/server/model-work-coordinator.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, it } from 'vitest';
+import { describe, expect, it } from 'vite-plus/test';
 
 import { ModelWorkCoordinator } from '../../packages/server/src/model-work-coordinator.js';
 

__test__/server/server-close.test.ts

@@ -1,6 +1,6 @@
 import { once } from 'node:events';
 import { request as httpRequest } from 'node:http';
-import type { AddressInfo } from 'node:net';
+import { connect, type AddressInfo } from 'node:net';
 
 import type { SessionCapableModel } from '@mlx-node/lm';
 import { activeSSEStreamCount, createServer, type ServerInstance } from '@mlx-node/server';
@@ -278,7 +278,15 @@ describe('close() on a quiet server', () => {
   it('stops the server from accepting new requests', async () => {
     const { instance, base } = await start();
     await instance.close({ timeoutMs: 1_000 });
-    await expect(fetch(`${base}/health`)).rejects.toThrow();
+    const connectionError = await new Promise<NodeJS.ErrnoException | null>((resolve) => {
+      const socket = connect(Number(new URL(base).port), '127.0.0.1');
+      socket.once('error', (error) => resolve(error as NodeJS.ErrnoException));
+      socket.once('connect', () => {
+        socket.destroy();
+        resolve(null);
+      });
+    });
+    expect(connectionError?.code).toBe('ECONNREFUSED');
   });
 });
 

__test__/tokenization/qwen3-tokenizer.test.ts

@@ -4,25 +4,16 @@
  * Comprehensive tests for the Rust-based Qwen3 tokenizer implementation
  */
 
-import { join } from 'node:path';
-import { fileURLToPath } from 'node:url';
-
 import { Qwen3Tokenizer, type ChatMessage } from '@mlx-node/core';
 import { describe, it, expect, beforeAll } from 'vite-plus/test';
 
+import { findTokenizerPath } from '../test-model-utils';
+
 describe('Qwen3Tokenizer', () => {
   let tokenizer: Qwen3Tokenizer;
 
-  const TOKENIZER_PATH = join(
-    fileURLToPath(import.meta.url),
-    '..',
-    '..',
-    '..',
-    '.cache/models/qwen3-0.6b-mlx-bf16/tokenizer.json',
-  );
-
   beforeAll(async () => {
-    tokenizer = await Qwen3Tokenizer.fromPretrained(TOKENIZER_PATH);
+    tokenizer = await Qwen3Tokenizer.fromPretrained(findTokenizerPath());
   });
 
   describe('Initialization', () => {