Repository navigation
llama.cpp POWER8 Port
LLM inference optimized for IBM POWER8 with VSX/AltiVec.
Features
- IBM MASS library integration (6-7x speedup)
- NUMA-aware memory management
- PSE vec_perm collapse optimization
- 576GB RAM support
Based on upstream llama.cpp with POWER8-specific optimizations.