JANGQ-AI/Nemotron-Cascade-2-30B-A3B-JANG_4M

Add vMLX app banner and runtime note to model card

Add vMLX app banner to model card

config.json: per-module quant overrides + rope_parameters (metadata bug fix)

capabilities: stamp reasoning/tool parser + cache_type for vmlx CapabilityDetector Tier-1

Add open-source links and provenance to README

Add @dealignai X/Twitter handle to README

Nemotron-Cascade-2 JANG_4M: 93% MMLU reasoning, 17 GB, beats MLX 4-bit

initial commit