D1A for MLX (Apple Silicon GPU)
D1A decision models as MLX exports, for Apple Silicon GPUs.
Text Classification • 5B • Updated • 182Note E2B-it QAT v0.7, MLX 4-bit on Google's QAT int4 grid: decisions, chat, photo/voice/video; v0.7 is better at English/Japanese dialogue acts (recommended E2B-it; v0.6 tagged)
mlx-community/d1a-e2b-it-qat-4bit
Text Classification • 5B • Updated • 420 • 2Note mlx-community mirror (byte-identical, same tags)
JohnP1/d1a-e4b-it-qat-mlx-q4
Text Classification • 7B • Updated • 128Note E4B-it QAT, MLX 4-bit on Google's QAT int4 grid: decisions, chat, photo/voice/video; level with v0.6 overall, typesafe −7.8
mlx-community/d1a-e4b-it-qat-4bit
Text Classification • 7B • Updated • 230Note mlx-community mirror (byte-identical)
JohnP1/d1a-e4b-mlx-q8
Text Classification • 7B • Updated • 82Note E4B, MLX 8-bit (v0.6)
JohnP1/d1a-e4b-it-mlx-q8
Text Classification • 7B • Updated • 77Note E4B-it, MLX 8-bit with photo/voice/video encoders (adapter merged: decisions only)
JohnP1/d1a-e4b-it-mlx-q4
Text Classification • 7B • Updated • 72Note E4B-it, MLX 4-bit with photo/voice/video encoders (adapter merged; moves more than 8-bit)
JohnP1/d1a-e2b-mlx-q8
Text Classification • 5B • Updated • 79 • 1Note E2B, MLX 8-bit
JohnP1/d1a-e2b-it-mlx-q4
Text Classification • 5B • Updated • 67 • 1Note E2B-it, MLX 4-bit, experimental: for the iOS proof of concept
JohnP1/d1a-e2b-it-mlx-q8
Text Classification • 5B • Updated • 56 • 1Note E2B-it, MLX 8-bit with photo/voice/video encoders, experimental
mlx-community/d1a-e4b-8bit
Text Classification • 7B • Updated • 73Note mlx-community mirror (byte-identical, v0.6)
mlx-community/d1a-e4b-it-8bit
Text Classification • 7B • Updated • 171Note mlx-community mirror (byte-identical, v0.6)
mlx-community/d1a-e2b-8bit
Text Classification • 5B • Updated • 93Note mlx-community mirror (byte-identical, v0.6)
mlx-community/d1a-e2b-it-8bit
Text Classification • 5B • Updated • 115Note mlx-community mirror (byte-identical, v0.6)