Hugging Face's logo
Activity Feed Request to join this org

AI & ML interests

None defined yet.

Recent Activity
yoshphysΒ  updated a model about 9 hours ago
mlx-community/Irodori-TTS-v4.1-Small-8bit
yoshphysΒ  published a model about 9 hours ago
mlx-community/Irodori-TTS-v4.1-Small-8bit
yoshphysΒ  updated a model about 9 hours ago
mlx-community/Irodori-TTS-v4.1-Small-fp16
View all activity

mlx-community 's collections 190

Qwen3.8
Qwen3.8 models converted to MLX format for Apple silicon.
North-Micro-Vision
MLX quantizations of CohereLabs/North-Micro-Vision-Instruct.
Community Contribution
A collection made to show a group of AI models hosted by community.
Laguna-S-2.1
MLX versions of Laguna-S-2.1
Real-ESRGAN (MLX)
Apple MLX fp16 ports of Real-ESRGAN super-resolution (RRDBNet + SRVGGNetCompact), 5 variants, BSD-3.
Abliterated Huihui LFM2.5-8B-A1B (MLX)
kinda of uncensored LFM-2.5-8B-1B-Activated-of-8B
Lens 3.8B (MLX)
Apple MLX conversions of microsoft/Lens β€” 3.8B text-to-image DiT (GPT-OSS features + FLUX.2 VAE) for Apple Silicon. bf16 + int4/int8.
LongCat-Video-Avatar 1.5 β€” MLX
Apple MLX port of Meituan's audio-driven video diffusion. Source + recipe: github.com/xocialize/longcat-avatar-mlx
MiMo-V2.5-ASR
by Xiaomi, converted to MLX
Gemma-4 Assistant (MTP)
Nvidia Nemotron-3-Nano-Omni
KittenTTS
All MLX conversions of KittenTTS (nano/micro/mini) across fp32, fp16, bf16, and 4/5/6/8-bit quantizations.
FunctionGemma
by Google Deepmind
Josiefied and Abliterated Models
πŸ’§LFM2-8B-A1B-MoE
Best in Class MoE, better than Qwen3. Optimised for Smaller devices sub 16 GB (M1/2/3/4) Apple Silicon.
Qwen3-Coder-MoE
πŸ’» Significant Performance: among open models on Agentic Coding, Agentic Browser-Use, and other foundational coding tasks, achieving ~Claude Sonnet.
Qwen3 Next
Alibaba's first hybrid model, designed to cut resources and speed things up.
Apertus
SwissAI's Apertus models that support 1k languages
Gemma 3n - Text Only (LM)
Google's Gemma 3n converted to MLX using mlx-lm
DeepSeek R1 0528
Gemma 3 DWQ
Gemma 3 distilled weight quantized (DWQ) models
Josiefied and Abliterated Qwen3
Abliterated, and further fine-tuned to be the most uncensored models available. Now in MLX
Gemma 3 QAT
Quantization Aware Trained (QAT) Gemma 3 checkpoints. The model preserves similar quality as half precision while using 3x less memory.
Kokoro TTS
Kokoro is an open-weight TTS model with 82 million parameters. Despite its lightweight architecture, it delivers amazing quality.
Josiefied and Abliterated Qwen2.5
The best uncensored models
Qwen2.5-Coder
Code-specific model series based on Qwen2.5
Qwen1.5
Qwen1.5 is the improved version of Qwen, the large language model series developed by Alibaba Cloud.
Mamba
Mamba is a new LLM architecture that integrates the Structured State Space sequence model to manage lengthy data sequences.
Nemotron-Parse
MLX conversions of NVIDIA Nemotron-Parse-2.0: first MLX support for the architecture (new port, issue #1865 / PR #1866).
Ministral-3 Base
MLX versions of Ministral-3 Base
Inkling MLX
MLX versions of Inkling 975B-A41B and 276B-A12B (Small) omni to text.
Gemma-4-12b-coder-fable5-composer2.5
MLX conversions of Gemma-4-12b-coder-fable5-composer2.5 for Apple Silicon Chips
Cocktail-Fork MRX (MLX)
MERL MRX ported to Apple MLX β€” 3-stem music/speech/sfx soundtrack separation. Numerically exact vs PyTorch. 4 variants.
Qwen 3.x MTP
MLX MTP drafter checkpoints for Qwen 3.x speculative decoding with mlx-vlm.
SongGeneration v2 MLX
Apple MLX checkpoints for Tencent SongGeneration v2 medium and large audiolm token generation.
MiniCPM-V 4.6
MLX variants of MiniCPM-V 4.6, 1.3B parameters (SigLIP2 400M vision encoder + Qwen3.5-0.8B LLM), repo: https://huggingface.co/openbmb/MiniCPM-V-4.6
UNCENSORED Qwen 3.6 27B
MedGemma-1.5
MedGemma-1.5 models in MLX format. See original repo: https://huggingface.co/google/medgemma-1.5-4b-it
Gabliterated v1
The next version of Abliteration
Olmo-3
Ai2's Olmo 3 model family of instruction and reasoning models.
ServiceNow-Apriel
Apriel-1.5-15b-Thinker is a multimodal reasoning model in ServiceNow’s Apriel SLM series which achieves competitive performance against models 10 time
SEA-LION
SEA-LION mlx models by AI Singapore.
EmbeddingGemma
BitNet 1.58
This collection houses BitNet-1.58, Falcon3-1.58 and Falcon-E quants.
Qwen3 DWQ Quants
High-quality 4-bit quants of the Qwen3 model family.
MedGemma
Collection of Gemma 3 variants for performance on medical text and image comprehension to accelerate building healthcare-based AI applications.
Parakeet
Nvidia's ASR models, now in MLX!
GLM4
The GLM-4 and Z1 series are powerful open-source language models excelling in reasoning, code, and complex tasks.
Gemma 3
A collection of lightweight, state-of-the-art open models built from the same research and technology that powers the Gemini 2.0 models
Qwen2.5-VL
Helium-1
Kyutai's Helium-1 2B Model, outperforming other state of the art small models.
Qwen2.5
The Qwen 2.5 models are a series of AI models trained on 18 trillion tokens, supporting 29 languages and offering advanced features such as instructio
Llama 3.1
Llama 3.2
Meta goes small with Llama3.2, both text only 1B and 3B, and the 11B Vision models.
Qwen3.8
Qwen3.8 models converted to MLX format for Apple silicon.
Nemotron-Parse
MLX conversions of NVIDIA Nemotron-Parse-2.0: first MLX support for the architecture (new port, issue #1865 / PR #1866).
North-Micro-Vision
MLX quantizations of CohereLabs/North-Micro-Vision-Instruct.
Community Contribution
A collection made to show a group of AI models hosted by community.
Laguna-S-2.1
MLX versions of Laguna-S-2.1
Ministral-3 Base
MLX versions of Ministral-3 Base
Inkling MLX
MLX versions of Inkling 975B-A41B and 276B-A12B (Small) omni to text.
Gemma-4-12b-coder-fable5-composer2.5
MLX conversions of Gemma-4-12b-coder-fable5-composer2.5 for Apple Silicon Chips
Real-ESRGAN (MLX)
Apple MLX fp16 ports of Real-ESRGAN super-resolution (RRDBNet + SRVGGNetCompact), 5 variants, BSD-3.
Cocktail-Fork MRX (MLX)
MERL MRX ported to Apple MLX β€” 3-stem music/speech/sfx soundtrack separation. Numerically exact vs PyTorch. 4 variants.
Abliterated Huihui LFM2.5-8B-A1B (MLX)
kinda of uncensored LFM-2.5-8B-1B-Activated-of-8B
Qwen 3.x MTP
MLX MTP drafter checkpoints for Qwen 3.x speculative decoding with mlx-vlm.
Lens 3.8B (MLX)
Apple MLX conversions of microsoft/Lens β€” 3.8B text-to-image DiT (GPT-OSS features + FLUX.2 VAE) for Apple Silicon. bf16 + int4/int8.
LongCat-Video-Avatar 1.5 β€” MLX
Apple MLX port of Meituan's audio-driven video diffusion. Source + recipe: github.com/xocialize/longcat-avatar-mlx
SongGeneration v2 MLX
Apple MLX checkpoints for Tencent SongGeneration v2 medium and large audiolm token generation.
MiMo-V2.5-ASR
by Xiaomi, converted to MLX
MiniCPM-V 4.6
MLX variants of MiniCPM-V 4.6, 1.3B parameters (SigLIP2 400M vision encoder + Qwen3.5-0.8B LLM), repo: https://huggingface.co/openbmb/MiniCPM-V-4.6
Gemma-4 Assistant (MTP)
UNCENSORED Qwen 3.6 27B
Nvidia Nemotron-3-Nano-Omni
KittenTTS
All MLX conversions of KittenTTS (nano/micro/mini) across fp32, fp16, bf16, and 4/5/6/8-bit quantizations.
MedGemma-1.5
MedGemma-1.5 models in MLX format. See original repo: https://huggingface.co/google/medgemma-1.5-4b-it
Gabliterated v1
The next version of Abliteration
FunctionGemma
by Google Deepmind
Josiefied and Abliterated Models
Olmo-3
Ai2's Olmo 3 model family of instruction and reasoning models.
πŸ’§LFM2-8B-A1B-MoE
Best in Class MoE, better than Qwen3. Optimised for Smaller devices sub 16 GB (M1/2/3/4) Apple Silicon.
ServiceNow-Apriel
Apriel-1.5-15b-Thinker is a multimodal reasoning model in ServiceNow’s Apriel SLM series which achieves competitive performance against models 10 time
Qwen3-Coder-MoE
πŸ’» Significant Performance: among open models on Agentic Coding, Agentic Browser-Use, and other foundational coding tasks, achieving ~Claude Sonnet.
Qwen3 Next
Alibaba's first hybrid model, designed to cut resources and speed things up.
SEA-LION
SEA-LION mlx models by AI Singapore.
EmbeddingGemma
Apertus
SwissAI's Apertus models that support 1k languages
Gemma 3n - Text Only (LM)
Google's Gemma 3n converted to MLX using mlx-lm
BitNet 1.58
This collection houses BitNet-1.58, Falcon3-1.58 and Falcon-E quants.
Qwen3 DWQ Quants
High-quality 4-bit quants of the Qwen3 model family.
DeepSeek R1 0528
MedGemma
Collection of Gemma 3 variants for performance on medical text and image comprehension to accelerate building healthcare-based AI applications.
Gemma 3 DWQ
Gemma 3 distilled weight quantized (DWQ) models
Parakeet
Nvidia's ASR models, now in MLX!
Josiefied and Abliterated Qwen3
Abliterated, and further fine-tuned to be the most uncensored models available. Now in MLX
GLM4
The GLM-4 and Z1 series are powerful open-source language models excelling in reasoning, code, and complex tasks.
Gemma 3 QAT
Quantization Aware Trained (QAT) Gemma 3 checkpoints. The model preserves similar quality as half precision while using 3x less memory.
Gemma 3
A collection of lightweight, state-of-the-art open models built from the same research and technology that powers the Gemini 2.0 models
Kokoro TTS
Kokoro is an open-weight TTS model with 82 million parameters. Despite its lightweight architecture, it delivers amazing quality.
Qwen2.5-VL
Helium-1
Kyutai's Helium-1 2B Model, outperforming other state of the art small models.
Josiefied and Abliterated Qwen2.5
The best uncensored models
Qwen2.5-Coder
Code-specific model series based on Qwen2.5
Qwen1.5
Qwen1.5 is the improved version of Qwen, the large language model series developed by Alibaba Cloud.
Qwen2.5
The Qwen 2.5 models are a series of AI models trained on 18 trillion tokens, supporting 29 languages and offering advanced features such as instructio
Mamba
Mamba is a new LLM architecture that integrates the Structured State Space sequence model to manage lengthy data sequences.
Llama 3.1
Llama 3.2
Meta goes small with Llama3.2, both text only 1B and 3B, and the 11B Vision models.