]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 502a400)
Retire the ggml_mul_mat() branch for transposed src0 (#500)
authorGeorgi Gerganov <redacted>
Sat, 25 Mar 2023 17:47:21 +0000 (19:47 +0200)
committerGitHub <redacted>
Sat, 25 Mar 2023 17:47:21 +0000 (19:47 +0200)
commitecbe466a364876927994e2f1ec14f4d82301d201
tree618c3e4e3f178f20bdc219e80b455ff45381ada5tree
parent502a400192013d3e95ed87b777e8fa3bec45713ccommit | diff
Retire the ggml_mul_mat() branch for transposed src0 (#500)

* Retire the ggml_mul_mat() for transposed src0

- It can always be made contiguous with ggml_cpy()
- The code is now simplified
- The results are deterministic in respect to num threads

* SIMD-ify dequantize_row_q4_0() for ARM_NEON (#502)

* Attempt to SIMD-ify dequantize_row_q4_0() for ARM_NEON

* Fix dequantization - forgot to interleave the quants
ggml.c diff | blob | history
Packaging of ggml-org/llama.cpp