Palmyra Mini MLX BF16 Overview
Writer's Palmyra Mini MLX BF16 is a 1.7 billion parameter language model built on the Qwen2 architecture, engineered for optimal performance on Apple Silicon (M1, M2, M3, M4 series) devices. It leverages the MLX framework and maintains full bfloat16 precision, ensuring high-quality inference. The model features a substantial 131,072-token context window, enabling it to process and understand very long inputs.
Key Capabilities
- Exceptional Mathematical Reasoning: Achieves strong scores on benchmarks like GSM8K (0.818) and MATH500 (0.818), indicating proficiency in grade-school and competition-level math problems.
- Complex Problem Solving: Demonstrates robust reasoning abilities, scoring 0.5259 on the Big-Bench Hard (BBH) benchmark.
- Apple Silicon Optimization: Specifically designed for efficient execution on Apple's M-series chips, requiring approximately 3.3GB of memory for weights.
- Long Context Handling: Supports a maximum position embedding of 131,072 tokens, suitable for tasks requiring extensive contextual understanding.
Good For
- Research and Development: Ideal for generative AI applications focused on mathematical and logical reasoning.
- Apple Silicon Users: Provides an optimized solution for developers and researchers working with Apple hardware.
- Tasks Requiring Deep Understanding: Suitable for problems that demand multi-step thought processes and extensive context.