llama.cpp
CUDA: Add BF16 path to CUBLAS and increase precision of FP16 path
#20078

Open

CUDA: Add BF16 path to CUBLAS and increase precision of FP16 path #20078

ORippler wants to merge 2 commits into ggml-org:master from ORippler:osimons/add_bf16_cublas_path

CUDA: Add BF16 CUBLAS path

5296977b

CUBLAS: Use FP32 accumulation also for VOLTA TCs

9c3032b7

github-actions added Nvidia GPU

github-actions added ggml

Reviewers

No reviews

Assignees

No one assigned

Labels

Nvidia GPU ggml

Milestone

No milestone