]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: a4b07c0)
`ggml_cuda_cpy` support for 4d tensors and float16->float32 upcasting (ggml/686)
authorJohn Balis <redacted>
Mon, 29 Jan 2024 12:37:33 +0000 (06:37 -0600)
committerGeorgi Gerganov <redacted>
Tue, 30 Jan 2024 14:20:25 +0000 (16:20 +0200)
commit625a699b5456994bc32a8093d53818f60ceda6d1
treea2803e2bd4787418016e906886c2266d6b45293ctree
parenta4b07c057a553b1ac253051efc3f040351e2eae1commit | diff
`ggml_cuda_cpy` support for 4d tensors and float16->float32 upcasting (ggml/686)

* added cuda float16->float32 upcasting to ggml_cuda_cpy

* added ability to copy 4d tensors with the cuda backend

* added tests for float16_>float32 upcast and 4d tensor cuda copys

* added 4d copy test for float32->float16 copy

* applied patch suggested by @iamlemec

* simplify cpy tests

---------

Co-authored-by: slaren <redacted>
ggml-cuda.cu diff | blob | history
tests/test-backend-ops.cpp diff | blob | history
Packaging of ggml-org/llama.cpp