Text-to-speech models quantized for mlx-audio; every tier passed a multilingual synthesis smoke gate (ASR round-trip CER) before upload.
Volodymyr IERMOLAIEV PRO
majentik
AI & ML interests
Bleeding-edge AI models and most prominent ML open-source projects.
Recent Activity
published a model 3 days ago
majentik/Swift-Qwen3.8-27B-MLX-3bit updated a model 3 days ago
majentik/Swift-Qwen3.8-27B-MLX-MXFP4 published a model 3 days ago
majentik/Swift-Qwen3.8-27B-MLX-MXFP4Organizations
Agentic 30B vision-language model from Meta, quantized for Apple Silicon with MLX.
Quantized GGUF and MLX packs of the Nemotron 3 model family (Nano, Nano-Omni, Super).
-
majentik/Nemotron-3-Nano-30B-A3B-RotorQuant-GGUF-IQ4_XS
Text Generation • 32B • Updated • 227 -
majentik/Nemotron-3-Nano-30B-A3B-RotorQuant-GGUF-Q2_K
Text Generation • 32B • Updated • 75 -
majentik/Nemotron-3-Nano-30B-A3B-RotorQuant-GGUF-Q3_K_M
Text Generation • 32B • Updated • 84 -
majentik/Nemotron-3-Nano-30B-A3B-RotorQuant-GGUF-Q5_K_M
Text Generation • 32B • Updated • 99
The verified, rebuilt gpt-oss-20b RotorQuant GGUF quant family (the retired 120b tombstones are excluded).
-
majentik/gpt-oss-20b-RotorQuant-GGUF-IQ4_XS
Text Generation • 21B • Updated • 480 -
majentik/gpt-oss-20b-RotorQuant-GGUF-Q2_K
Text Generation • 21B • Updated • 105 -
majentik/gpt-oss-20b-RotorQuant-GGUF-Q3_K_M
Text Generation • 21B • Updated • 112 -
majentik/gpt-oss-20b-RotorQuant-GGUF-Q4_K_M
Text Generation • 21B • Updated • 205
Automatic speech recognition and transcription/diarization MLX packs that run natively on Apple Silicon.
-
majentik/cohere-transcribe-arabic-07-2026-MLX-4bit
Automatic Speech Recognition • 2B • Updated • 37 -
majentik/cohere-transcribe-arabic-07-2026-MLX-6bit
Automatic Speech Recognition • 2B • Updated • 28 -
majentik/cohere-transcribe-arabic-07-2026-MLX-8bit
Automatic Speech Recognition • 2B • Updated • 110 -
majentik/MERaLiON-3-3B-ASR-MLX-8bit
Automatic Speech Recognition • 3B • Updated • 35
Quantized GGUF and MLX packs of the Qwen 3.5 model family, including the 397B-A17B MoE.
-
majentik/Qwen3.5-27B-RotorQuant-GGUF-IQ4_XS
Text Generation • 27B • Updated • 276 -
majentik/Qwen3.5-27B-RotorQuant-GGUF-Q2_K
Text Generation • 27B • Updated • 95 • 2 -
majentik/Qwen3.5-27B-RotorQuant-GGUF-Q3_K_M
Text Generation • 27B • Updated • 100 -
majentik/Qwen3.5-27B-RotorQuant-GGUF-Q4_K_M
Text Generation • 27B • Updated • 113
Quantized GGUF and MLX packs of the Gemma 4 family (dense and MoE variants) for local inference.
-
majentik/gemma-4-26B-A4B-it-RotorQuant-GGUF-IQ4_XS
Image-Text-to-Text • 25B • Updated • 365 -
majentik/gemma-4-26B-A4B-it-RotorQuant-GGUF-Q2_K
Image-Text-to-Text • 25B • Updated • 204 -
majentik/gemma-4-26B-A4B-it-RotorQuant-GGUF-Q3_K_M
Image-Text-to-Text • 25B • Updated • 160 -
majentik/gemma-4-26B-A4B-it-RotorQuant-GGUF-Q4_K_M
Image-Text-to-Text • 25B • Updated • 300 • 1
UEmbed multimodal embedding models running dense text embeddings on Apple Silicon via MLX.
Qwen3-Embedding model packs in ONNX, GGUF, and MLX formats for retrieval and embedding workloads.
-
majentik/Qwen3-Embedding-0.6B-GGUF-IQ4_XS
Feature Extraction • 0.6B • Updated • 66 • 1 -
majentik/Qwen3-Embedding-0.6B-GGUF-Q4_K_M
Feature Extraction • 0.6B • Updated • 75 -
majentik/Qwen3-Embedding-0.6B-GGUF-Q5_K_M
Feature Extraction • 0.6B • Updated • 98 -
majentik/Qwen3-Embedding-0.6B-GGUF-Q8_0
Feature Extraction • 0.6B • Updated • 50
Session-verified MLX packs that passed a smoke gate against a reference implementation before publication.
-
majentik/MERaLiON-3-10B-MLX-8bit
Automatic Speech Recognition • 10B • Updated • 6 -
majentik/MERaLiON-3-10B-MLX-4bit
Automatic Speech Recognition • 10B • Updated • 12 -
majentik/MERaLiON-3-3B-ASR-MLX-8bit
Automatic Speech Recognition • 3B • Updated • 35 -
majentik/MOSS-Transcribe-Diarize-MLX-8bit
Automatic Speech Recognition • 0.9B • Updated • 95 • 1
OCR and document-understanding MLX packs, including Unlimited-OCR and the GELab-Zero preview family.
-
majentik/GELab-Zero-4B-preview-Sico-Evolution-MLX-4bit
Image-Text-to-Text • 4B • Updated • 6 -
majentik/GELab-Zero-4B-preview-Sico-Evolution-MLX-6bit
Image-Text-to-Text • 4B • Updated • 8 -
majentik/GELab-Zero-4B-preview-Sico-Evolution-MLX-8bit
Image-Text-to-Text • 4B • Updated • 4 -
majentik/Unlimited-OCR-MLX-6bit
Image-Text-to-Text • 3B • Updated • 33
MERaLiON speech and ASR model packs (MLX, GGUF, quantized variants) for Apple Silicon.
-
majentik/MERaLiON-3-10B-MLX
Automatic Speech Recognition • Updated • 21 -
majentik/MERaLiON-3-10B-MLX-4bit
Automatic Speech Recognition • 10B • Updated • 12 -
majentik/MERaLiON-3-10B-MLX-8bit
Automatic Speech Recognition • 10B • Updated • 6 -
majentik/MERaLiON-3-10B-RotorQuant-MLX-2bit
Automatic Speech Recognition • Updated • 19
Quantized GGUF, MLX, and FP8/FP4 packs of the Qwen 3.6 model family.
Text-to-speech models quantized for mlx-audio; every tier passed a multilingual synthesis smoke gate (ASR round-trip CER) before upload.
UEmbed multimodal embedding models running dense text embeddings on Apple Silicon via MLX.
Agentic 30B vision-language model from Meta, quantized for Apple Silicon with MLX.
Qwen3-Embedding model packs in ONNX, GGUF, and MLX formats for retrieval and embedding workloads.
-
majentik/Qwen3-Embedding-0.6B-GGUF-IQ4_XS
Feature Extraction • 0.6B • Updated • 66 • 1 -
majentik/Qwen3-Embedding-0.6B-GGUF-Q4_K_M
Feature Extraction • 0.6B • Updated • 75 -
majentik/Qwen3-Embedding-0.6B-GGUF-Q5_K_M
Feature Extraction • 0.6B • Updated • 98 -
majentik/Qwen3-Embedding-0.6B-GGUF-Q8_0
Feature Extraction • 0.6B • Updated • 50
Quantized GGUF and MLX packs of the Nemotron 3 model family (Nano, Nano-Omni, Super).
-
majentik/Nemotron-3-Nano-30B-A3B-RotorQuant-GGUF-IQ4_XS
Text Generation • 32B • Updated • 227 -
majentik/Nemotron-3-Nano-30B-A3B-RotorQuant-GGUF-Q2_K
Text Generation • 32B • Updated • 75 -
majentik/Nemotron-3-Nano-30B-A3B-RotorQuant-GGUF-Q3_K_M
Text Generation • 32B • Updated • 84 -
majentik/Nemotron-3-Nano-30B-A3B-RotorQuant-GGUF-Q5_K_M
Text Generation • 32B • Updated • 99
Session-verified MLX packs that passed a smoke gate against a reference implementation before publication.
-
majentik/MERaLiON-3-10B-MLX-8bit
Automatic Speech Recognition • 10B • Updated • 6 -
majentik/MERaLiON-3-10B-MLX-4bit
Automatic Speech Recognition • 10B • Updated • 12 -
majentik/MERaLiON-3-3B-ASR-MLX-8bit
Automatic Speech Recognition • 3B • Updated • 35 -
majentik/MOSS-Transcribe-Diarize-MLX-8bit
Automatic Speech Recognition • 0.9B • Updated • 95 • 1
The verified, rebuilt gpt-oss-20b RotorQuant GGUF quant family (the retired 120b tombstones are excluded).
-
majentik/gpt-oss-20b-RotorQuant-GGUF-IQ4_XS
Text Generation • 21B • Updated • 480 -
majentik/gpt-oss-20b-RotorQuant-GGUF-Q2_K
Text Generation • 21B • Updated • 105 -
majentik/gpt-oss-20b-RotorQuant-GGUF-Q3_K_M
Text Generation • 21B • Updated • 112 -
majentik/gpt-oss-20b-RotorQuant-GGUF-Q4_K_M
Text Generation • 21B • Updated • 205
OCR and document-understanding MLX packs, including Unlimited-OCR and the GELab-Zero preview family.
-
majentik/GELab-Zero-4B-preview-Sico-Evolution-MLX-4bit
Image-Text-to-Text • 4B • Updated • 6 -
majentik/GELab-Zero-4B-preview-Sico-Evolution-MLX-6bit
Image-Text-to-Text • 4B • Updated • 8 -
majentik/GELab-Zero-4B-preview-Sico-Evolution-MLX-8bit
Image-Text-to-Text • 4B • Updated • 4 -
majentik/Unlimited-OCR-MLX-6bit
Image-Text-to-Text • 3B • Updated • 33
Automatic speech recognition and transcription/diarization MLX packs that run natively on Apple Silicon.
-
majentik/cohere-transcribe-arabic-07-2026-MLX-4bit
Automatic Speech Recognition • 2B • Updated • 37 -
majentik/cohere-transcribe-arabic-07-2026-MLX-6bit
Automatic Speech Recognition • 2B • Updated • 28 -
majentik/cohere-transcribe-arabic-07-2026-MLX-8bit
Automatic Speech Recognition • 2B • Updated • 110 -
majentik/MERaLiON-3-3B-ASR-MLX-8bit
Automatic Speech Recognition • 3B • Updated • 35
MERaLiON speech and ASR model packs (MLX, GGUF, quantized variants) for Apple Silicon.
-
majentik/MERaLiON-3-10B-MLX
Automatic Speech Recognition • Updated • 21 -
majentik/MERaLiON-3-10B-MLX-4bit
Automatic Speech Recognition • 10B • Updated • 12 -
majentik/MERaLiON-3-10B-MLX-8bit
Automatic Speech Recognition • 10B • Updated • 6 -
majentik/MERaLiON-3-10B-RotorQuant-MLX-2bit
Automatic Speech Recognition • Updated • 19
Quantized GGUF and MLX packs of the Qwen 3.5 model family, including the 397B-A17B MoE.
-
majentik/Qwen3.5-27B-RotorQuant-GGUF-IQ4_XS
Text Generation • 27B • Updated • 276 -
majentik/Qwen3.5-27B-RotorQuant-GGUF-Q2_K
Text Generation • 27B • Updated • 95 • 2 -
majentik/Qwen3.5-27B-RotorQuant-GGUF-Q3_K_M
Text Generation • 27B • Updated • 100 -
majentik/Qwen3.5-27B-RotorQuant-GGUF-Q4_K_M
Text Generation • 27B • Updated • 113
Quantized GGUF, MLX, and FP8/FP4 packs of the Qwen 3.6 model family.
Quantized GGUF and MLX packs of the Gemma 4 family (dense and MoE variants) for local inference.
-
majentik/gemma-4-26B-A4B-it-RotorQuant-GGUF-IQ4_XS
Image-Text-to-Text • 25B • Updated • 365 -
majentik/gemma-4-26B-A4B-it-RotorQuant-GGUF-Q2_K
Image-Text-to-Text • 25B • Updated • 204 -
majentik/gemma-4-26B-A4B-it-RotorQuant-GGUF-Q3_K_M
Image-Text-to-Text • 25B • Updated • 160 -
majentik/gemma-4-26B-A4B-it-RotorQuant-GGUF-Q4_K_M
Image-Text-to-Text • 25B • Updated • 300 • 1