Open Issues Need Help
View All on GitHub Beat 19.3 t/s decode at 86k context on 4x P100 (open challenge) about 2 months ago
help wanted
poisonxa16/pxq_llama.cpp poisonxa16/pxq_llama.cpp The codec + kernel pack for cards with HBM2 and no DP4A: Pascal P100 and Volta V100, fully GPU-resident, 84 architectures, dense and MoE
Open project →
20
The codec + kernel pack for cards with HBM2 and no DP4A: Pascal P100 and Volta V100, fully GPU-resident, 84 architectures, dense and MoE
C++
#cuda#gguf#inference#llama-cpp#moe#p100#pascal#quantization#v100#volta