diff options
| author | Iwan Kawrakow <iwan.kawrakow@gmail.com> | 2024-06-18 13:56:16 +0300 |
|---|---|---|
| committer | Iwan Kawrakow <iwan.kawrakow@gmail.com> | 2024-06-22 12:02:52 +0300 |
| commit | 01ea9a862d4afb73f936de8f4ef46401ce11b596 (patch) | |
| tree | 892dd475ddf47c9ba401859b285b4b517d36135e /ggml-cuda/common.cuh | |
| parent | 2998ca9b14d9b2d4b184cf6d923cea8b07a6320a (diff) | |
Bitnet(2.25 bpw): CUDA
We get PP-512 = 9600 t/s, TG-128 = 234 t/s
(but we need to use 8 CPU threads, else results are lower,
so clearly there is something being computed on the CPU).
PP-512 is very close to PP-512(fp16) = 9800 t/s
Diffstat (limited to 'ggml-cuda/common.cuh')
| -rw-r--r-- | ggml-cuda/common.cuh | 7 |
1 files changed, 7 insertions, 0 deletions
diff --git a/ggml-cuda/common.cuh b/ggml-cuda/common.cuh index c21513e6..892fd5a6 100644 --- a/ggml-cuda/common.cuh +++ b/ggml-cuda/common.cuh @@ -630,6 +630,13 @@ struct ggml_cuda_type_traits<GGML_TYPE_IQ1_BN> { }; template<> +struct ggml_cuda_type_traits<GGML_TYPE_IQ2_BN> { + static constexpr int qk = QK_IQ1BN; + static constexpr int qr = QR1_BN; + static constexpr int qi = QI1_BN; +}; + +template<> struct ggml_cuda_type_traits<GGML_TYPE_IQ4_NL> { static constexpr int qk = QK4_NL; static constexpr int qr = QR4_NL; |