SDAR Language Models & Quantization

11 repos

Specialized quantized variants of large language models using the SDAR (likely a custom architecture or quantization method) framework, with model sizes ranging from 4B to 8B parameters and bit-widths of 32 and 64. The cluster focuses on different model configurations and thinking/reasoning variants, stored in safetensors format for efficient distribution and inference. These repositories represent a systematic exploration of trade-offs between model capacity, precision, and computational efficiency for deployment-ready LLM implementations.