matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-11 00:07:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggufv3

Browse Source 
ggufv3 adds support for big endianness, mainly for s390x architecture.
while that's not currently supported for ollama, the change is simple.

loosen version check to be more forward compatible. unless specified,
gguf versions other v1 will be decoded into v2.
This commit is contained in:
Michael Yang
2023-10-23 09:33:13 -07:00
parent ccff9ca09c
commit 125d0a013a
2 changed files with 36 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
llm/ggml.go
View File

@@ -175,7 +175,8 @@ const (
// Magic constant for `ggla` files (LoRA adapter).
FILE_MAGIC_GGLA = 0x67676C61
// Magic constant for `gguf` files (versioned, gguf)
FILE_MAGIC_GGUF = 0x46554747
FILE_MAGIC_GGUF_LE = 0x46554747
FILE_MAGIC_GGUF_BE = 0x47475546
)
func DecodeGGML(r io.ReadSeeker) (*GGML, error) {
@@ -191,8 +192,10 @@ func DecodeGGML(r io.ReadSeeker) (*GGML, error) {
ggml.container = &containerGGJT{}
case FILE_MAGIC_GGLA:
ggml.container = &containerLORA{}
case FILE_MAGIC_GGUF:
ggml.container = &containerGGUF{}
case FILE_MAGIC_GGUF_LE:
ggml.container = &containerGGUF{bo: binary.LittleEndian}
case FILE_MAGIC_GGUF_BE:
ggml.container = &containerGGUF{bo: binary.BigEndian}
default:
return nil, errors.New("invalid file magic")
}