Oscilla/OSLO-1.5B-MLX-4bit

OSLO 1.5B MLX

OSLO 1.5B MLX is the MLX-optimized version of Oscilla's first lightweight language model, designed for fast, private AI inference on Apple Silicon.

Model Details

Model Description

OSLO 1.5B MLX is an MLX-formatted version of OSLO 1.5B, optimized for efficient inference on Apple Silicon. It is designed for fast, local conversations, writing assistance, coding, and research while taking advantage of Apple's unified memory architecture.

• Developed by: Oscilla
• Model type: Causal language model
• Language(s): English
• License: More information needed
• Base model: Qwen/Qwen2.5-1.5B-Instruct
• Converted from: Oscilla/OSLO-1.5B-BETA

Uses

Direct Use

OSLO 1.5B MLX is intended for local inference using the MLX framework on Apple Silicon devices.

Common use cases include:

• Conversational AI
• Writing assistance
• Coding support
• Research
• On-device AI applications
• Apple platform development

Downstream Use

This model is designed for applications built with MLX, including native macOS and iOS experiences through the Oscilla platform.

Out-of-Scope Use

This model should not be relied upon for legal, medical, financial, or other safety-critical decisions. Outputs should always be verified where accuracy is important.

Bias, Risks, and Limitations

Like other language models, OSLO 1.5B MLX may:

• Produce inaccurate information
• Reflect biases present in training data
• Struggle with complex reasoning tasks
• Have limited world knowledge beyond its training cutoff

Recommendations

Always verify important information and use appropriate human oversight for high-impact applications.

Getting Started

Install MLX-LM:

pip install mlx-lm

Run the model:

mlx_lm.generate \
    --model Oscilla/OSLO-1.5B-MLX \
    --prompt "What is AI?"

Training Details

Training Data

OSLO 1.5B MLX is a converted version of OSLO 1.5B and shares the same fine-tuning process and training data.

Training Procedure

• Base Model: Qwen/Qwen2.5-1.5B-Instruct
• Fine-tuning: LoRA
• Merge: Standalone Hugging Face model
• Conversion: MLX
• Quantization: MLX Quantized

Technical Specifications

Architecture

• Decoder-only Transformer
• Instruction tuned
• MLX optimized
• Apple Silicon optimized

More Information

OSLO is Oscilla's family of language models focused on private, local AI experiences for Apple devices. The MLX versions are optimized specifically for efficient inference on Apple Silicon using Apple's MLX framework.

Model Card Authors

Oscilla

Model Card Contact

Oscilla