ferrox-core
Core tensor ops, RoPE, GQA, and KV cache for Ferrox.
WeightMatrix, RMSNorm, quantized matmul, grouped-query attention, and
optional CUDA/Metal apply_gpu dispatch when those features are enabled.
Core tensor ops, RoPE, GQA, and KV cache for Ferrox.
WeightMatrix, RMSNorm, quantized matmul, grouped-query attention, and
optional CUDA/Metal apply_gpu dispatch when those features are enabled.