cb42e607c5
Previously, some costly things were causing the loading of GGUF files
and their metadata and tensor information to be VERY slow:
* Too many allocations when decoding strings
* Hitting disk for each read of each key and value, resulting in a
not-okay amount of syscalls/disk I/O.
The show API is now down to 33ms from 800ms+ for llama3 on a macbook pro
m3.
This commit also prevents collecting large arrays of values when
decoding GGUFs (if desired). When such keys are encountered, their
values are null, and are encoded as such in JSON.
Also, this fixes a broken test that was not encoding valid GGUF.
34 lines
590 B
Go
34 lines
590 B
Go
package bufioutil
|
|
|
|
import (
|
|
"bufio"
|
|
"io"
|
|
)
|
|
|
|
type BufferedSeeker struct {
|
|
rs io.ReadSeeker
|
|
br *bufio.Reader
|
|
}
|
|
|
|
func NewBufferedSeeker(rs io.ReadSeeker, size int) *BufferedSeeker {
|
|
return &BufferedSeeker{
|
|
rs: rs,
|
|
br: bufio.NewReaderSize(rs, size),
|
|
}
|
|
}
|
|
|
|
func (b *BufferedSeeker) Read(p []byte) (int, error) {
|
|
return b.br.Read(p)
|
|
}
|
|
|
|
func (b *BufferedSeeker) Seek(offset int64, whence int) (int64, error) {
|
|
if whence == io.SeekCurrent {
|
|
offset -= int64(b.br.Buffered())
|
|
}
|
|
n, err := b.rs.Seek(offset, whence)
|
|
if err != nil {
|
|
return 0, err
|
|
}
|
|
b.br.Reset(b.rs)
|
|
return n, nil
|
|
}
|