Gradio API migration, training pipeline, news page, and UI improvements

- Migrate backend from REST API to Gradio @gradio/client for generation - Fix Gradio parameter alignment (positions 36-49) for reference/cover audio - Add LoRA training pipeline with dataset upload, preprocessing, and export - Add News page with dismiss/restore and GitHub star button - Add localization info icon in Settings language section - Fix upload audio URL prefix, add missing MIME types - Add training API routes and Python preprocess script - Update i18n with news keys for all languages
2026-02-09 22:30:15 +02:00
parent f42fde9b40
commit 565faacb7b
23 changed files with 3145 additions and 236 deletions
@@ -106,6 +106,7 @@ export interface Song {
  user_id?: string;
  created_at: string;
  creator?: string;
+  creator_avatar?: string;
  ditModel?: string;
  generation_params?: any;
 }
@@ -311,11 +312,14 @@ export interface GenerationParams {

 export interface GenerationJob {
  jobId: string;
+  id?: string;
  status: 'pending' | 'queued' | 'running' | 'succeeded' | 'failed';
  queuePosition?: number;
  etaSeconds?: number;
  progress?: number;
  stage?: string;
+  params?: any;
+  created_at?: string;
  result?: {
    audioUrls: string[];
    bpm?: number;
@@ -375,6 +379,13 @@ export const generateApi = {
    error?: string;
  }> => api('/api/generate/format', { method: 'POST', body: params, token }),

+  // Random description from Gradio's example library
+  getRandomDescription: (token: string): Promise<{
+    description: string;
+    instrumental: boolean;
+    vocalLanguage: string;
+  }> => api('/api/generate/random-description', { token }),
+
  // LoRA Inference (requires ACE-Step training fork)
  loadLora: (params: {
    lora_path: string;
@@ -393,6 +404,20 @@ export const generateApi = {
    message: string;
    scale: number;
  }> => api('/api/lora/scale', { method: 'POST', body: params, token }),
+
+  toggleLora: (params: {
+    enabled: boolean;
+  }, token: string): Promise<{
+    message: string;
+    active: boolean;
+  }> => api('/api/lora/toggle', { method: 'POST', body: params, token }),
+
+  getLoraStatus: (token: string): Promise<{
+    loaded: boolean;
+    active: boolean;
+    scale: number;
+    path: string;
+  }> => api('/api/lora/status', { token }),
 };

 // Users API
@@ -532,3 +557,238 @@ export const contactApi = {
  submit: (data: ContactFormData): Promise<{ success: boolean; message: string; id: string }> =>
    api('/api/contact', { method: 'POST', body: data }),
 };
+
+// Training API (LoRA fine-tuning via Gradio)
+export interface TrainingSample {
+  audio: unknown;
+  filename: string;
+  caption: string;
+  genre: string;
+  promptOverride: string;
+  lyrics: string;
+  bpm: number;
+  key: string;
+  timeSignature: string;
+  duration: number;
+  language: string;
+  instrumental: boolean;
+  rawLyrics?: string;
+}
+
+export interface DatasetSettings {
+  datasetName: string;
+  customTag: string;
+  tagPosition: 'prepend' | 'append' | 'replace';
+  allInstrumental: boolean;
+  genreRatio: number;
+}
+
+export interface TrainingParams {
+  tensorDir?: string;
+  rank?: number;
+  alpha?: number;
+  dropout?: number;
+  learningRate?: number;
+  epochs?: number;
+  batchSize?: number;
+  gradientAccumulation?: number;
+  saveEvery?: number;
+  shift?: number;
+  seed?: number;
+  outputDir?: string;
+  resumeCheckpoint?: string | null;
+}
+
+// Helper: build proxy URL for training audio files
+export function getTrainingAudioUrl(audioPath: unknown, token?: string): string | undefined {
+  if (!audioPath) return undefined;
+
+  // Handle Gradio FileData objects
+  if (typeof audioPath === 'object' && audioPath !== null) {
+    const fd = audioPath as Record<string, unknown>;
+    if (fd.url && typeof fd.url === 'string') return fd.url;
+    if (fd.path && typeof fd.path === 'string') {
+      return `${API_BASE}/api/training/audio?path=${encodeURIComponent(fd.path)}`;
+    }
+    return undefined;
+  }
+
+  // Handle absolute path string
+  if (typeof audioPath === 'string') {
+    if (audioPath.startsWith('http://') || audioPath.startsWith('https://') || audioPath.startsWith('/audio/')) {
+      return audioPath;
+    }
+    return `${API_BASE}/api/training/audio?path=${encodeURIComponent(audioPath)}`;
+  }
+
+  return undefined;
+}
+
+export const trainingApi = {
+  // Upload audio files for a dataset
+  uploadAudio: async (files: File[], datasetName: string, token: string): Promise<{
+    files: Array<{ filename: string; originalName: string; size: number; path: string }>;
+    uploadDir: string;
+    count: number;
+  }> => {
+    const formData = new FormData();
+    formData.append('datasetName', datasetName);
+    for (const file of files) {
+      formData.append('audio', file);
+    }
+    const response = await fetch(`${API_BASE}/api/training/upload-audio`, {
+      method: 'POST',
+      headers: { 'Authorization': `Bearer ${token}` },
+      body: formData,
+    });
+    if (!response.ok) {
+      const error = await response.json().catch(() => ({ error: 'Upload failed' }));
+      throw new Error(error.error || 'Upload failed');
+    }
+    return response.json();
+  },
+
+  // Build dataset JSON from uploaded audio files
+  buildDataset: (params: {
+    datasetName: string;
+    customTag?: string;
+    tagPosition?: string;
+    allInstrumental?: boolean;
+  }, token: string): Promise<{
+    status: string;
+    dataframe: unknown;
+    sampleCount: number;
+    sample: TrainingSample;
+    settings: DatasetSettings;
+    datasetPath: string;
+  }> => api('/api/training/build-dataset', { method: 'POST', body: params, token }),
+
+  // Scan directory for audio files (Node.js implementation)
+  scanDirectory: (params: {
+    audioDir: string;
+    datasetName?: string;
+    customTag?: string;
+    tagPosition?: string;
+    allInstrumental?: boolean;
+  }, token: string): Promise<{
+    status: string;
+    dataframe: unknown;
+    sampleCount: number;
+    audioDir: string;
+  }> => api('/api/training/scan-directory', { method: 'POST', body: params, token }),
+
+  // Auto-label dataset samples (requires model loaded in Gradio)
+  autoLabel: (params: {
+    skipMetas?: boolean;
+    formatLyrics?: boolean;
+    transcribeLyrics?: boolean;
+    onlyUnlabeled?: boolean;
+  }, token: string): Promise<{
+    dataframe?: unknown;
+    status: string;
+    error?: string;
+    hint?: string;
+  }> => api('/api/training/auto-label', { method: 'POST', body: params, token }),
+
+  // Initialize model for training (requires Gradio)
+  initModel: (params: {
+    checkpoint?: string;
+    configPath?: string;
+    device?: string;
+    initLlm?: boolean;
+    lmModelPath?: string;
+    backend?: string;
+    useFlashAttention?: boolean;
+    offloadToCpu?: boolean;
+    offloadDitToCpu?: boolean;
+    compileModel?: boolean;
+    quantization?: boolean;
+  }, token: string): Promise<{
+    status: string;
+    modelReady?: boolean;
+    error?: string;
+    hint?: string;
+  }> => api('/api/training/init-model', { method: 'POST', body: params, token }),
+
+  // List available checkpoints
+  getCheckpoints: (token: string): Promise<{
+    checkpoints: string[];
+    configs: string[];
+  }> => api('/api/training/checkpoints', { token }),
+
+  // List LoRA training checkpoints
+  getLoraCheckpoints: (dir: string, token: string): Promise<{
+    checkpoints: string[];
+    outputDir: string;
+  }> => api(`/api/training/lora-checkpoints?dir=${encodeURIComponent(dir)}`, { token }),
+
+  // Preprocess dataset to tensors
+  preprocess: (params: {
+    datasetPath: string;
+    outputDir?: string;
+  }, token: string): Promise<{
+    status: string;
+    message?: string;
+    output_files?: number;
+  }> => api('/api/training/preprocess', { method: 'POST', body: params, token }),
+
+  loadDataset: (datasetPath: string, token: string): Promise<{
+    status: string;
+    dataframe: unknown;
+    sampleCount: number;
+    sample: TrainingSample;
+    settings: DatasetSettings;
+  }> => api('/api/training/load-dataset', { method: 'POST', body: { datasetPath }, token }),
+
+  getSamplePreview: (idx: number, token: string): Promise<TrainingSample> =>
+    api(`/api/training/sample-preview?idx=${idx}`, { token }),
+
+  saveSample: (params: {
+    sampleIdx: number;
+    caption: string;
+    genre: string;
+    promptOverride: string;
+    lyrics: string;
+    bpm: number;
+    key: string;
+    timeSignature: string;
+    language: string;
+    instrumental: boolean;
+  }, token: string): Promise<{ dataframe: unknown; status: string }> =>
+    api('/api/training/save-sample', { method: 'POST', body: params, token }),
+
+  updateSettings: (params: {
+    customTag: string;
+    tagPosition: string;
+    allInstrumental: boolean;
+    genreRatio: number;
+  }, token: string): Promise<{ success: boolean }> =>
+    api('/api/training/update-settings', { method: 'POST', body: params, token }),
+
+  saveDataset: (params: {
+    savePath?: string;
+    datasetName?: string;
+  }, token: string): Promise<{ status: string; path: string }> =>
+    api('/api/training/save-dataset', { method: 'POST', body: params, token }),
+
+  loadTensors: (tensorDir: string, token: string): Promise<{ status: string }> =>
+    api('/api/training/load-tensors', { method: 'POST', body: { tensorDir }, token }),
+
+  startTraining: (params: TrainingParams, token: string): Promise<{
+    progress: string;
+    log: string;
+    metrics: unknown;
+  }> => api('/api/training/start', { method: 'POST', body: params, token }),
+
+  stopTraining: (token: string): Promise<{ status: string }> =>
+    api('/api/training/stop', { method: 'POST', token }),
+
+  exportLora: (params: {
+    exportPath?: string;
+    loraOutputDir?: string;
+  }, token: string): Promise<{ status: string }> =>
+    api('/api/training/export', { method: 'POST', body: params, token }),
+
+  importDataset: (datasetType: string, token: string): Promise<{ status: string }> =>
+    api('/api/training/import-dataset', { method: 'POST', body: { datasetType }, token }),
+};