Retire the ggml_mul_mat() branch for transposed src0 (#500)

Browse source

* 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

...

This commit is contained in:

Georgi Gerganov 2023-03-25 19:47:21 +02:00 committed by GitHub

parent 502a400192

commit ecbe466a36

No known key found for this signature in database

GPG key ID: 4AEE18F83AFDEB23

1 changed files with 277 additions and 756 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Show stats Download patch file Download diff file Expand all files Collapse all files

1033

ggml.c

View file

File diff suppressed because it is too large Load diff