2023-07-21 20:33:56 +00:00
|
|
|
package llm
|
|
|
|
|
|
|
|
import (
|
|
|
|
"encoding/binary"
|
|
|
|
"errors"
|
|
|
|
"io"
|
|
|
|
)
|
|
|
|
|
2023-09-12 17:01:20 +00:00
|
|
|
type GGML struct {
|
|
|
|
container
|
|
|
|
model
|
2023-11-24 19:57:20 +00:00
|
|
|
|
|
|
|
Size int64
|
2023-09-12 17:01:20 +00:00
|
|
|
}
|
2023-07-21 20:33:56 +00:00
|
|
|
|
|
|
|
const (
|
2023-09-12 17:01:20 +00:00
|
|
|
fileTypeF32 uint32 = iota
|
|
|
|
fileTypeF16
|
|
|
|
fileTypeQ4_0
|
|
|
|
fileTypeQ4_1
|
|
|
|
fileTypeQ4_1_F16
|
|
|
|
fileTypeQ8_0 uint32 = iota + 2
|
|
|
|
fileTypeQ5_0
|
|
|
|
fileTypeQ5_1
|
|
|
|
fileTypeQ2_K
|
|
|
|
fileTypeQ3_K_S
|
|
|
|
fileTypeQ3_K_M
|
|
|
|
fileTypeQ3_K_L
|
|
|
|
fileTypeQ4_K_S
|
|
|
|
fileTypeQ4_K_M
|
|
|
|
fileTypeQ5_K_S
|
|
|
|
fileTypeQ5_K_M
|
|
|
|
fileTypeQ6_K
|
2024-02-21 00:06:29 +00:00
|
|
|
fileTypeIQ2_XXS
|
|
|
|
fileTypeIQ2_XS
|
|
|
|
fileTypeQ2_K_S
|
|
|
|
fileTypeQ3_K_XS
|
|
|
|
fileTypeIQ3_XXS
|
2023-07-21 20:33:56 +00:00
|
|
|
)
|
|
|
|
|
2023-09-12 17:01:20 +00:00
|
|
|
func fileType(fileType uint32) string {
|
|
|
|
switch fileType {
|
|
|
|
case fileTypeF32:
|
|
|
|
return "F32"
|
|
|
|
case fileTypeF16:
|
|
|
|
return "F16"
|
|
|
|
case fileTypeQ4_0:
|
|
|
|
return "Q4_0"
|
|
|
|
case fileTypeQ4_1:
|
|
|
|
return "Q4_1"
|
|
|
|
case fileTypeQ4_1_F16:
|
|
|
|
return "Q4_1_F16"
|
|
|
|
case fileTypeQ8_0:
|
|
|
|
return "Q8_0"
|
|
|
|
case fileTypeQ5_0:
|
|
|
|
return "Q5_0"
|
|
|
|
case fileTypeQ5_1:
|
|
|
|
return "Q5_1"
|
|
|
|
case fileTypeQ2_K:
|
|
|
|
return "Q2_K"
|
|
|
|
case fileTypeQ3_K_S:
|
|
|
|
return "Q3_K_S"
|
|
|
|
case fileTypeQ3_K_M:
|
|
|
|
return "Q3_K_M"
|
|
|
|
case fileTypeQ3_K_L:
|
|
|
|
return "Q3_K_L"
|
|
|
|
case fileTypeQ4_K_S:
|
|
|
|
return "Q4_K_S"
|
|
|
|
case fileTypeQ4_K_M:
|
|
|
|
return "Q4_K_M"
|
|
|
|
case fileTypeQ5_K_S:
|
|
|
|
return "Q5_K_S"
|
|
|
|
case fileTypeQ5_K_M:
|
|
|
|
return "Q5_K_M"
|
|
|
|
case fileTypeQ6_K:
|
|
|
|
return "Q6_K"
|
2024-02-21 00:06:29 +00:00
|
|
|
case fileTypeIQ2_XXS:
|
|
|
|
return "IQ2_XXS"
|
|
|
|
case fileTypeIQ2_XS:
|
|
|
|
return "IQ2_XS"
|
|
|
|
case fileTypeQ2_K_S:
|
|
|
|
return "Q2_K_S"
|
|
|
|
case fileTypeQ3_K_XS:
|
|
|
|
return "Q3_K_XS"
|
|
|
|
case fileTypeIQ3_XXS:
|
|
|
|
return "IQ3_XXS"
|
2023-08-17 18:37:27 +00:00
|
|
|
default:
|
2023-10-03 02:52:25 +00:00
|
|
|
return "unknown"
|
2023-08-17 18:37:27 +00:00
|
|
|
}
|
|
|
|
}
|
2023-07-21 20:33:56 +00:00
|
|
|
|
2023-08-17 18:37:27 +00:00
|
|
|
type model interface {
|
2023-09-12 17:01:20 +00:00
|
|
|
ModelFamily() string
|
|
|
|
ModelType() string
|
|
|
|
FileType() string
|
2024-01-08 21:42:00 +00:00
|
|
|
NumLayers() uint32
|
|
|
|
NumGQA() uint32
|
|
|
|
NumEmbed() uint32
|
|
|
|
NumHead() uint32
|
|
|
|
NumHeadKv() uint32
|
2024-01-12 22:54:01 +00:00
|
|
|
NumCtx() uint32
|
2023-07-21 20:33:56 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
type container interface {
|
|
|
|
Name() string
|
2024-03-09 20:28:36 +00:00
|
|
|
Decode(io.ReadSeeker) (model, error)
|
2023-07-21 20:33:56 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
const (
|
2023-09-07 17:55:37 +00:00
|
|
|
// Magic constant for `ggml` files (unversioned).
|
2023-07-21 20:33:56 +00:00
|
|
|
FILE_MAGIC_GGML = 0x67676d6c
|
2023-09-07 17:55:37 +00:00
|
|
|
// Magic constant for `ggml` files (versioned, ggmf).
|
2023-07-21 20:33:56 +00:00
|
|
|
FILE_MAGIC_GGMF = 0x67676d66
|
2023-09-07 17:55:37 +00:00
|
|
|
// Magic constant for `ggml` files (versioned, ggjt).
|
2023-07-21 20:33:56 +00:00
|
|
|
FILE_MAGIC_GGJT = 0x67676a74
|
2023-09-07 17:55:37 +00:00
|
|
|
// Magic constant for `ggla` files (LoRA adapter).
|
2023-07-21 20:33:56 +00:00
|
|
|
FILE_MAGIC_GGLA = 0x67676C61
|
2023-09-07 17:55:37 +00:00
|
|
|
// Magic constant for `gguf` files (versioned, gguf)
|
2023-10-23 16:33:13 +00:00
|
|
|
FILE_MAGIC_GGUF_LE = 0x46554747
|
|
|
|
FILE_MAGIC_GGUF_BE = 0x47475546
|
2023-07-21 20:33:56 +00:00
|
|
|
)
|
|
|
|
|
2023-11-24 18:58:09 +00:00
|
|
|
var ErrUnsupportedFormat = errors.New("unsupported model format")
|
|
|
|
|
2024-03-09 20:28:36 +00:00
|
|
|
func DecodeGGML(rs io.ReadSeeker) (*GGML, error) {
|
2023-11-24 19:57:20 +00:00
|
|
|
var magic uint32
|
2024-03-09 20:28:36 +00:00
|
|
|
if err := binary.Read(rs, binary.LittleEndian, &magic); err != nil {
|
2023-11-24 19:57:20 +00:00
|
|
|
return nil, err
|
|
|
|
}
|
|
|
|
|
|
|
|
var c container
|
|
|
|
switch magic {
|
2023-11-24 18:58:09 +00:00
|
|
|
case FILE_MAGIC_GGML, FILE_MAGIC_GGMF, FILE_MAGIC_GGJT:
|
|
|
|
return nil, ErrUnsupportedFormat
|
2023-07-21 20:33:56 +00:00
|
|
|
case FILE_MAGIC_GGLA:
|
2024-03-08 23:38:53 +00:00
|
|
|
c = &ContainerGGLA{}
|
2023-10-23 16:33:13 +00:00
|
|
|
case FILE_MAGIC_GGUF_LE:
|
2024-03-07 05:01:51 +00:00
|
|
|
c = &ContainerGGUF{ByteOrder: binary.LittleEndian}
|
2023-10-23 16:33:13 +00:00
|
|
|
case FILE_MAGIC_GGUF_BE:
|
2024-03-07 05:01:51 +00:00
|
|
|
c = &ContainerGGUF{ByteOrder: binary.BigEndian}
|
2023-07-21 20:33:56 +00:00
|
|
|
default:
|
|
|
|
return nil, errors.New("invalid file magic")
|
|
|
|
}
|
|
|
|
|
2024-03-09 20:28:36 +00:00
|
|
|
model, err := c.Decode(rs)
|
2024-03-08 23:38:53 +00:00
|
|
|
if errors.Is(err, io.EOF) {
|
|
|
|
// noop
|
|
|
|
} else if err != nil {
|
2023-07-21 20:33:56 +00:00
|
|
|
return nil, err
|
|
|
|
}
|
|
|
|
|
2024-03-09 20:28:36 +00:00
|
|
|
offset, err := rs.Seek(0, io.SeekCurrent)
|
|
|
|
if err != nil {
|
|
|
|
return nil, err
|
|
|
|
}
|
|
|
|
|
2023-07-21 20:33:56 +00:00
|
|
|
// final model type
|
2023-11-24 19:57:20 +00:00
|
|
|
return &GGML{
|
|
|
|
container: c,
|
|
|
|
model: model,
|
2024-03-09 20:28:36 +00:00
|
|
|
Size: offset,
|
2023-11-24 19:57:20 +00:00
|
|
|
}, nil
|
|
|
|
}
|