master-20a8549
Add RWKV v5.1 and v5.2 support (#145) * Reformat CMakeLists and docs * Add RWKV v5.1 and v5.2 support * Remove sequence_length-based heuristic for allocating ggml context * Set GGML_CUDA_MMV_Y to 2 by default * Clarify comment; update ggml * Make some operations inplace * Make some operations inplace * Add late_abort option for tests * Increase thresholds * Increase thresholds * Update README.md