vikramlingam/Ensemble-MLX-Vault

Ensemble MLX Vault

⚠️ Important Legal & Usage Notice

This repository is a Private Personal Storage Vault

Sourcing & Ownership

• Not an Original Creator: The models stored in this repository are NOT created, trained, or owned by the owner of this repository.
• Sourced from MLX-Community: All models are redistributed from the MLX-Community organization on Hugging Face.
• Purpose: This repository serves as a optimized, mobile-ready mirror of specific 4-bit quantized versions to ensure consistent and fast access.

Compliance & Licenses

The weights in this repository are subject to the original licenses provided by their respective authors (Meta, Google, Mistral, Microsoft, Sarvam AI, Alibaba/Qwen, etc.).

• Users must adhere to the individual model licenses (e.g., Llama 3.2 Community License, Apache 2.0, MIT) as defined by the original creators.
• This repository is provided "as is" without any warranties.

Repository Overview

This vault contains 30 model snapshots, each optimized for Apple Silicon via the MLX framework.

Selection Criteria:

1. Size: Strictly under 4 Billion parameters.
2. Quantization: 4-bit (Safetensors).
3. Weight Limit: Every model folder is < 3.5GB.

Key Included Series:

• Qwen 3.5 / 2.5 (VL, Coder, and Base)
• Gemma 3 / 2 (Vision and Text)
• Llama 3.2 / DeepSeek R1 (Instruct and Reasoning)
• Sarvam AI (Indic and Translation)
• Phi 4 / 3.5 / 2
• SmolLM2 / Danube / Falcon 3

Sync Status

Last Synchronized: March 10, 2026. Contact: via Private Message on HF for access inquiries for the Ensemble AI team.