nf4-quantization

A 4-bit NormalFloat (NF4) quantized version of InstaDeep's Nucleotide Transformer 2.5B Multi-Species model, optimized for significantly lower GPU memory consumption while preserving embedding quality and inference performance.

Fine tuned qwen llm using qloRa for commentary generation and generating audio using edge-tts.

🚀 3.3B Mamba-MoE running on Mac Mini M4 16GB! Breakthrough deployment of Concept models on consumer Apple Silicon via NF4 quantization & QLoRA. Pushing limits of on-device AI.

spec-rama is a parameter-efficient fine-tuning (PEFT) framework that combines Weight Permutation (Greedy TSP / Bipartite) with Spectral Core Adaptation (DCT, FWHT, DWT Wavelet) and RAMA Multiplicative-Additive Modulation.