Voice-to-text dictation app with local (Nvidia Parakeet/Whisper) and cloud models (BYOK). Privacy-first and available cross-platform.
JavaScript aianthropiccross-platformgeminigroqlinuxmacosnvidiaopen-sourceopenaiparakeetspeech-to-texttranscribewhisperwindows
feat(models): add Gemma 4 QAT local models (#1268)
Alcahest Malus Dienheime1eca67ee6a5
diff --git a/src/locales/de/translation.json b/src/locales/de/translation.jsonindex 86877f6b4..873228871 100644--- a/src/locales/de/translation.json+++ b/src/locales/de/translation.json@@ -2003,9 +2003,13 @@"openai_oss_gpt_oss_20b_mxfp4": "Open-Weight-Modell von OpenAI für Consumer-Hardware","gemma_gemma_3_4b_it_q4_k_m": "Googles Open-Weight-Modell, gutes Gleichgewicht aus Geschwindigkeit und Qualität","gemma_gemma_4_31b_it_q4_k_m": "Googles größtes Gemma 4, beste Qualität",+ "gemma_gemma_4_31b_it_qat_q4_0": "Offizieller QAT-Build von Google, kleiner bei gleicher Qualität","gemma_gemma_4_26b_a4b_it_q4_k_m": "Googles MoE Gemma 4, schnell mit 4B aktiven Parametern",+ "gemma_gemma_4_26b_a4b_it_qat_q4_0": "Offizieller QAT-MoE-Build von Google, kleiner bei gleicher Qualität","gemma_gemma_4_e4b_it_q4_k_m": "Googles neuestes multimodales Modell, starkes Reasoning",+ "gemma_gemma_4_e4b_it_qat_q4_0": "Offizieller QAT-Build von Google, starkes Reasoning, kleinere Größe","gemma_gemma_4_e2b_it_q4_k_m": "Kompaktes Gemma 4 von Google, schnell und effizient",+ "gemma_gemma_4_e2b_it_qat_q4_0": "Kompakter offizieller QAT-Build von Google, Qualität erhalten","gemma_gemma_3_12b_it_q4_k_m": "Googles Mittelklasse-Modell, starkes Reasoning","gemma_gemma_3_1b_it_q4_k_m": "Ultraschnelles Google-Modell, ideal für kurze Diktat-Bereinigung","liquidai_lfm2_5_1_2b_instruct_q4_k_m": "Effizientes On-Device-Modell von Liquid AI, hervorragende Anweisungsbefolgung",diff --git a/src/locales/en/translation.json b/src/locales/en/translation.jsonindex fbdf427b0..8380e6055 100644--- a/src/locales/en/translation.json+++ b/src/locales/en/translation.json@@ -2095,9 +2095,13 @@"openai_oss_gpt_oss_20b_mxfp4": "OpenAI's open-weight model for consumer hardware","gemma_gemma_3_4b_it_q4_k_m": "Google's open-weight model, great balance of speed and quality","gemma_gemma_4_31b_it_q4_k_m": "Google's largest Gemma 4, best quality",+ "gemma_gemma_4_31b_it_qat_q4_0": "Google's official QAT build, smaller with quality preserved","gemma_gemma_4_26b_a4b_it_q4_k_m": "Google's MoE Gemma 4, fast with 4B active params",+ "gemma_gemma_4_26b_a4b_it_qat_q4_0": "Google's official QAT MoE build, smaller with quality preserved","gemma_gemma_4_e4b_it_q4_k_m": "Google's latest multimodal, strong reasoning",+ "gemma_gemma_4_e4b_it_qat_q4_0": "Google's official QAT build, strong reasoning, smaller size","gemma_gemma_4_e2b_it_q4_k_m": "Google's compact Gemma 4, fast and efficient",+ "gemma_gemma_4_e2b_it_qat_q4_0": "Google's official compact QAT build, quality preserved","gemma_gemma_3_12b_it_q4_k_m": "Google's mid-range model, strong reasoning","gemma_gemma_3_1b_it_q4_k_m": "Ultra-fast Google model, best for short dictation cleanup","liquidai_lfm2_5_1_2b_instruct_q4_k_m": "Liquid AI's efficient on-device model, excellent instruction following",diff --git a/src/locales/es/translation.json b/src/locales/es/translation.jsonindex f7d224dba..a868c9a00 100644--- a/src/locales/es/translation.json+++ b/src/locales/es/translation.json@@ -2051,9 +2051,13 @@"openai_oss_gpt_oss_20b_mxfp4": "Modelo open-weight de OpenAI para hardware de consumo","gemma_gemma_3_4b_it_q4_k_m": "Modelo open-weight de Google, excelente equilibrio entre velocidad y calidad","gemma_gemma_4_31b_it_q4_k_m": "El mayor Gemma 4 de Google, mejor calidad",+ "gemma_gemma_4_31b_it_qat_q4_0": "Build QAT oficial de Google, más pequeño con la calidad intacta","gemma_gemma_4_26b_a4b_it_q4_k_m": "Gemma 4 MoE de Google, rápido con 4B parámetros activos",+ "gemma_gemma_4_26b_a4b_it_qat_q4_0": "Build QAT MoE oficial de Google, más pequeño con la calidad intacta","gemma_gemma_4_e4b_it_q4_k_m": "Último multimodal de Google, razonamiento avanzado",+ "gemma_gemma_4_e4b_it_qat_q4_0": "Build QAT oficial de Google, razonamiento avanzado en menos espacio","gemma_gemma_4_e2b_it_q4_k_m": "Gemma 4 compacto de Google, rápido y eficiente",+ "gemma_gemma_4_e2b_it_qat_q4_0": "Build QAT compacto oficial de Google, calidad preservada","gemma_gemma_3_12b_it_q4_k_m": "Modelo de gama media de Google, razonamiento sólido","gemma_gemma_3_1b_it_q4_k_m": "Modelo Google ultrarrápido, ideal para limpieza de dictado corto","liquidai_lfm2_5_1_2b_instruct_q4_k_m": "Modelo eficiente de Liquid AI, excelente seguimiento de instrucciones",diff --git a/src/locales/fr/translation.json b/src/locales/fr/translation.jsonindex 54de337e3..a8eab89cf 100644--- a/src/locales/fr/translation.json+++ b/src/locales/fr/translation.json@@ -2051,9 +2051,13 @@"openai_oss_gpt_oss_20b_mxfp4": "Modèle open-weight d'OpenAI pour matériel grand public","gemma_gemma_3_4b_it_q4_k_m": "Modèle open-weight de Google, excellent équilibre vitesse/qualité","gemma_gemma_4_31b_it_q4_k_m": "Le plus grand Gemma 4 de Google, meilleure qualité",+ "gemma_gemma_4_31b_it_qat_q4_0": "Build QAT officiel de Google, plus léger à qualité égale","gemma_gemma_4_26b_a4b_it_q4_k_m": "Gemma 4 MoE de Google, rapide avec 4B paramètres actifs",+ "gemma_gemma_4_26b_a4b_it_qat_q4_0": "Build QAT MoE officiel de Google, plus léger à qualité égale","gemma_gemma_4_e4b_it_q4_k_m": "Dernier multimodal de Google, raisonnement avancé",+ "gemma_gemma_4_e4b_it_qat_q4_0": "Build QAT officiel de Google, raisonnement avancé en moins d'espace","gemma_gemma_4_e2b_it_q4_k_m": "Gemma 4 compact de Google, rapide et efficace",+ "gemma_gemma_4_e2b_it_qat_q4_0": "Build QAT compact officiel de Google, qualité préservée","gemma_gemma_3_12b_it_q4_k_m": "Modèle milieu de gamme de Google, raisonnement solide","gemma_gemma_3_1b_it_q4_k_m": "Modèle Google ultra-rapide, idéal pour le nettoyage de courtes dictées","liquidai_lfm2_5_1_2b_instruct_q4_k_m": "Modèle efficace de Liquid AI, excellent suivi des instructions",diff --git a/src/locales/it/translation.json b/src/locales/it/translation.jsonindex c7abc93a4..c0387103c 100644--- a/src/locales/it/translation.json+++ b/src/locales/it/translation.json@@ -2003,9 +2003,13 @@"openai_oss_gpt_oss_20b_mxfp4": "Modello open-weight di OpenAI per hardware consumer","gemma_gemma_3_4b_it_q4_k_m": "Modello open-weight di Google, ottimo equilibrio tra velocità e qualità","gemma_gemma_4_31b_it_q4_k_m": "Il più grande Gemma 4 di Google, migliore qualità",+ "gemma_gemma_4_31b_it_qat_q4_0": "Build QAT ufficiale di Google, più leggero a parità di qualità","gemma_gemma_4_26b_a4b_it_q4_k_m": "Gemma 4 MoE di Google, veloce con 4B parametri attivi",+ "gemma_gemma_4_26b_a4b_it_qat_q4_0": "Build QAT MoE ufficiale di Google, più leggero a parità di qualità","gemma_gemma_4_e4b_it_q4_k_m": "Ultimo multimodale di Google, ragionamento avanzato",+ "gemma_gemma_4_e4b_it_qat_q4_0": "Build QAT ufficiale di Google, ragionamento avanzato in meno spazio","gemma_gemma_4_e2b_it_q4_k_m": "Gemma 4 compatto di Google, veloce ed efficiente",+ "gemma_gemma_4_e2b_it_qat_q4_0": "Build QAT compatta ufficiale di Google, qualità preservata","gemma_gemma_3_12b_it_q4_k_m": "Modello di fascia media di Google, ragionamento solido","gemma_gemma_3_1b_it_q4_k_m": "Modello Google ultraveloce, ideale per la pulizia di brevi dettature","liquidai_lfm2_5_1_2b_instruct_q4_k_m": "Modello efficiente di Liquid AI, eccellente nel seguire le istruzioni",diff --git a/src/locales/ja/translation.json b/src/locales/ja/translation.jsonindex 7160283a0..f2d852cb5 100644--- a/src/locales/ja/translation.json+++ b/src/locales/ja/translation.json@@ -2003,9 +2003,13 @@"openai_oss_gpt_oss_20b_mxfp4": "一般消費者向けハードウェア用 OpenAI オープンウェイトモデル","gemma_gemma_3_4b_it_q4_k_m": "Google のオープンウェイトモデル、速度と品質の優れたバランス","gemma_gemma_4_31b_it_q4_k_m": "Google 最大の Gemma 4、最高品質",+ "gemma_gemma_4_31b_it_qat_q4_0": "Google 公式の QAT ビルド、品質を保ちながら軽量","gemma_gemma_4_26b_a4b_it_q4_k_m": "Google の MoE Gemma 4、4B アクティブパラメータで高速",+ "gemma_gemma_4_26b_a4b_it_qat_q4_0": "Google 公式の QAT MoE ビルド、品質を保ちながら軽量","gemma_gemma_4_e4b_it_q4_k_m": "Google の最新マルチモーダルモデル、高度な推論",+ "gemma_gemma_4_e4b_it_qat_q4_0": "Google 公式の QAT ビルド、高度な推論をより小さく","gemma_gemma_4_e2b_it_q4_k_m": "Google のコンパクトな Gemma 4、高速で効率的",+ "gemma_gemma_4_e2b_it_qat_q4_0": "Google 公式のコンパクトな QAT ビルド、品質を維持","gemma_gemma_3_12b_it_q4_k_m": "Google の中型モデル、強力な推論","gemma_gemma_3_1b_it_q4_k_m": "超高速 Google モデル、短い口述のクリーンアップに最適","liquidai_lfm2_5_1_2b_instruct_q4_k_m": "Liquid AIの高効率オンデバイスモデル、優れた指示追従性",diff --git a/src/locales/pt/translation.json b/src/locales/pt/translation.jsonindex 292fce248..88b42a558 100644--- a/src/locales/pt/translation.json+++ b/src/locales/pt/translation.json@@ -1982,9 +1982,13 @@"openai_oss_gpt_oss_20b_mxfp4": "Modelo open-weight da OpenAI para hardware de consumo","gemma_gemma_3_4b_it_q4_k_m": "Modelo open-weight do Google, ótimo equilíbrio entre velocidade e qualidade","gemma_gemma_4_31b_it_q4_k_m": "O maior Gemma 4 do Google, melhor qualidade",+ "gemma_gemma_4_31b_it_qat_q4_0": "Build QAT oficial do Google, menor com a qualidade preservada","gemma_gemma_4_26b_a4b_it_q4_k_m": "Gemma 4 MoE do Google, rápido com 4B parâmetros ativos",+ "gemma_gemma_4_26b_a4b_it_qat_q4_0": "Build QAT MoE oficial do Google, menor com a qualidade preservada","gemma_gemma_4_e4b_it_q4_k_m": "Último multimodal do Google, raciocínio avançado",+ "gemma_gemma_4_e4b_it_qat_q4_0": "Build QAT oficial do Google, raciocínio avançado em menos espaço","gemma_gemma_4_e2b_it_q4_k_m": "Gemma 4 compacto do Google, rápido e eficiente",+ "gemma_gemma_4_e2b_it_qat_q4_0": "Build QAT compacto oficial do Google, qualidade preservada","gemma_gemma_3_12b_it_q4_k_m": "Modelo intermediário do Google, raciocínio sólido","gemma_gemma_3_1b_it_q4_k_m": "Modelo Google ultrarrápido, ideal para limpeza de ditados curtos","liquidai_lfm2_5_1_2b_instruct_q4_k_m": "Modelo eficiente da Liquid AI, excelente em seguir instruções",diff --git a/src/locales/ru/translation.json b/src/locales/ru/translation.jsonindex aea7d3853..dd0221248 100644--- a/src/locales/ru/translation.json+++ b/src/locales/ru/translation.json@@ -2003,9 +2003,13 @@"openai_oss_gpt_oss_20b_mxfp4": "Открытая модель OpenAI для потребительского оборудования","gemma_gemma_3_4b_it_q4_k_m": "Открытая модель Google, отличный баланс скорости и качества","gemma_gemma_4_31b_it_q4_k_m": "Крупнейшая Gemma 4 от Google, лучшее качество",+ "gemma_gemma_4_31b_it_qat_q4_0": "Официальная сборка QAT от Google, меньше при сохранении качества","gemma_gemma_4_26b_a4b_it_q4_k_m": "Gemma 4 MoE от Google, быстрая с 4B активными параметрами",+ "gemma_gemma_4_26b_a4b_it_qat_q4_0": "Официальная сборка QAT MoE от Google, меньше при сохранении качества","gemma_gemma_4_e4b_it_q4_k_m": "Новейшая мультимодальная модель Google, продвинутое рассуждение",+ "gemma_gemma_4_e4b_it_qat_q4_0": "Официальная сборка QAT от Google, продвинутое рассуждение в меньшем размере","gemma_gemma_4_e2b_it_q4_k_m": "Компактная Gemma 4 от Google, быстрая и эффективная",+ "gemma_gemma_4_e2b_it_qat_q4_0": "Компактная официальная сборка QAT от Google, качество сохранено","gemma_gemma_3_12b_it_q4_k_m": "Средняя модель Google, сильное рассуждение","gemma_gemma_3_1b_it_q4_k_m": "Сверхбыстрая модель Google, лучшая для очистки коротких диктовок","liquidai_lfm2_5_1_2b_instruct_q4_k_m": "Эффективная локальная модель Liquid AI, отлично следует инструкциям",diff --git a/src/locales/zh-CN/translation.json b/src/locales/zh-CN/translation.jsonindex bf4b88457..3272bf91f 100644--- a/src/locales/zh-CN/translation.json+++ b/src/locales/zh-CN/translation.json@@ -2003,9 +2003,13 @@"openai_oss_gpt_oss_20b_mxfp4": "OpenAI 开源模型,适用于消费级硬件","gemma_gemma_3_4b_it_q4_k_m": "谷歌开放权重模型,速度与质量的绝佳平衡","gemma_gemma_4_31b_it_q4_k_m": "Google 最大的 Gemma 4,最佳质量",+ "gemma_gemma_4_31b_it_qat_q4_0": "Google 官方 QAT 构建,体积更小且质量不减","gemma_gemma_4_26b_a4b_it_q4_k_m": "Google MoE Gemma 4,4B 活跃参数,速度快",+ "gemma_gemma_4_26b_a4b_it_qat_q4_0": "Google 官方 QAT MoE 构建,体积更小且质量不减","gemma_gemma_4_e4b_it_q4_k_m": "Google 最新多模态模型,强大推理能力",+ "gemma_gemma_4_e4b_it_qat_q4_0": "Google 官方 QAT 构建,强大推理能力,体积更小","gemma_gemma_4_e2b_it_q4_k_m": "Google 紧凑型 Gemma 4,快速高效",+ "gemma_gemma_4_e2b_it_qat_q4_0": "Google 官方紧凑型 QAT 构建,质量不减","gemma_gemma_3_12b_it_q4_k_m": "Google 中型模型,强推理能力","gemma_gemma_3_1b_it_q4_k_m": "超快谷歌模型,最适合短听写清理","liquidai_lfm2_5_1_2b_instruct_q4_k_m": "Liquid AI 高效端侧模型,指令遵循出色",diff --git a/src/locales/zh-TW/translation.json b/src/locales/zh-TW/translation.jsonindex 531bbde85..11a27411b 100644--- a/src/locales/zh-TW/translation.json+++ b/src/locales/zh-TW/translation.json@@ -2003,9 +2003,13 @@"openai_oss_gpt_oss_20b_mxfp4": "OpenAI 開放權重模型,適用於消費級硬體","gemma_gemma_3_4b_it_q4_k_m": "Google 開放權重模型,速度與品質的絕佳平衡","gemma_gemma_4_31b_it_q4_k_m": "Google 最大的 Gemma 4,最佳品質",+ "gemma_gemma_4_31b_it_qat_q4_0": "Google 官方 QAT 建置,體積更小且品質不減","gemma_gemma_4_26b_a4b_it_q4_k_m": "Google MoE Gemma 4,4B 活躍參數,速度快",+ "gemma_gemma_4_26b_a4b_it_qat_q4_0": "Google 官方 QAT MoE 建置,體積更小且品質不減","gemma_gemma_4_e4b_it_q4_k_m": "Google 最新多模態模型,強大推理能力",+ "gemma_gemma_4_e4b_it_qat_q4_0": "Google 官方 QAT 建置,強大推理能力,體積更小","gemma_gemma_4_e2b_it_q4_k_m": "Google 緊湊型 Gemma 4,快速高效",+ "gemma_gemma_4_e2b_it_qat_q4_0": "Google 官方緊湊型 QAT 建置,品質不減","gemma_gemma_3_12b_it_q4_k_m": "Google 中型模型,強推理能力","gemma_gemma_3_1b_it_q4_k_m": "超快 Google 模型,最適合短語音清理","liquidai_lfm2_5_1_2b_instruct_q4_k_m": "Liquid AI 高效端側模型,指令遵循出色",diff --git a/src/models/modelRegistryData.json b/src/models/modelRegistryData.jsonindex 4620f7224..f9b2b5b1a 100644--- a/src/models/modelRegistryData.json+++ b/src/models/modelRegistryData.json@@ -978,6 +978,18 @@"hfRepo": "bartowski/google_gemma-4-31B-it-GGUF","descriptionKey": "models.descriptions.local.gemma_gemma_4_31b_it_q4_k_m"},+ {+ "id": "gemma-4-31b-it-qat-q4_0",+ "name": "Gemma 4 31B QAT",+ "size": "17.7GB",+ "sizeBytes": 17651001568,+ "description": "Google's official QAT build, smaller with quality preserved",+ "fileName": "gemma-4-31B_q4_0-it.gguf",+ "quantization": "q4_0",+ "contextLength": 262144,+ "hfRepo": "google/gemma-4-31B-it-qat-q4_0-gguf",+ "descriptionKey": "models.descriptions.local.gemma_gemma_4_31b_it_qat_q4_0"+ },{"id": "gemma-4-26b-a4b-it-q4_k_m","name": "Gemma 4 26B MoE",@@ -990,6 +1002,18 @@"hfRepo": "bartowski/google_gemma-4-26B-A4B-it-GGUF","descriptionKey": "models.descriptions.local.gemma_gemma_4_26b_a4b_it_q4_k_m"},+ {+ "id": "gemma-4-26b-a4b-it-qat-q4_0",+ "name": "Gemma 4 26B MoE QAT",+ "size": "14.4GB",+ "sizeBytes": 14439363584,+ "description": "Google's official QAT MoE build, smaller with quality preserved",+ "fileName": "gemma-4-26B_q4_0-it.gguf",+ "quantization": "q4_0",+ "contextLength": 262144,+ "hfRepo": "google/gemma-4-26B-A4B-it-qat-q4_0-gguf",+ "descriptionKey": "models.descriptions.local.gemma_gemma_4_26b_a4b_it_qat_q4_0"+ },{"id": "gemma-4-e4b-it-q4_k_m","name": "Gemma 4 E4B",@@ -1000,9 +1024,21 @@"quantization": "q4_k_m","contextLength": 262144,"hfRepo": "bartowski/google_gemma-4-E4B-it-GGUF",- "recommended": true,"descriptionKey": "models.descriptions.local.gemma_gemma_4_e4b_it_q4_k_m"},+ {+ "id": "gemma-4-e4b-it-qat-q4_0",+ "name": "Gemma 4 E4B QAT",+ "size": "5.2GB",+ "sizeBytes": 5154941280,+ "description": "Google's official QAT build, strong reasoning, smaller size",+ "fileName": "gemma-4-E4B_q4_0-it.gguf",+ "quantization": "q4_0",+ "contextLength": 262144,+ "hfRepo": "google/gemma-4-E4B-it-qat-q4_0-gguf",+ "recommended": true,+ "descriptionKey": "models.descriptions.local.gemma_gemma_4_e4b_it_qat_q4_0"+ },{"id": "gemma-4-e2b-it-q4_k_m","name": "Gemma 4 E2B",@@ -1015,6 +1051,18 @@"hfRepo": "bartowski/google_gemma-4-E2B-it-GGUF","descriptionKey": "models.descriptions.local.gemma_gemma_4_e2b_it_q4_k_m"},+ {+ "id": "gemma-4-e2b-it-qat-q4_0",+ "name": "Gemma 4 E2B QAT",+ "size": "3.3GB",+ "sizeBytes": 3349516256,+ "description": "Google's official compact QAT build, quality preserved",+ "fileName": "gemma-4-E2B_q4_0-it.gguf",+ "quantization": "q4_0",+ "contextLength": 262144,+ "hfRepo": "google/gemma-4-E2B-it-qat-q4_0-gguf",+ "descriptionKey": "models.descriptions.local.gemma_gemma_4_e2b_it_qat_q4_0"+ },{"id": "gemma-3-12b-it-q4_k_m","name": "Gemma 3 12B",