Popular repositories Loading

1. mud-puppy mud-puppy Public

   ROCm-first LLM fine-tuning framework — LoRA, QLoRA, DPO/GRPO, GPTQ, ZeRO-Offload. No bitsandbytes dependency.

   Python

2. pensive pensive Public

   Hierarchical context retrieval for LLMs — L1/L2/L3 cache tiers, spreading activation, FAISS+BM25 hybrid search, 97.9% accuracy at sub-30ms

   Python

3. aegis aegis Public

   AI cognitive architecture: custom Triton INT4 inference kernels, hierarchical memory (Pensive), MoE routing, constitutional alignment — built on AMD ROCm

   HIP