Merge pull request #3664 from ollama/mxyng/fix-padding-2

fix padding to only return padding
This commit is contained in:
Michael Yang 2024-04-17 15:57:40 -07:00 committed by GitHub
commit 05e9424824
No known key found for this signature in database
GPG key ID: B5690EEEBB952194

View file

@ -248,13 +248,17 @@ func (llm *gguf) Decode(rs io.ReadSeeker) error {
} }
padding := llm.padding(offset, int64(alignment)) padding := llm.padding(offset, int64(alignment))
if _, err := rs.Seek(padding-offset, io.SeekCurrent); err != nil { if _, err := rs.Seek(padding, io.SeekCurrent); err != nil {
return err return err
} }
for _, tensor := range llm.tensors { for _, tensor := range llm.tensors {
padded := (int64(tensor.size()) + int64(alignment) - 1) & ^(int64(alignment) - 1) if _, err := rs.Seek(int64(tensor.size()), io.SeekCurrent); err != nil {
if _, err := rs.Seek(padded, io.SeekCurrent); err != nil { return err
}
padding := llm.padding(int64(tensor.size()), int64(alignment))
if _, err := rs.Seek(padding, io.SeekCurrent); err != nil {
return err return err
} }
} }
@ -623,8 +627,9 @@ func (llm *gguf) Encode(ws io.WriteSeeker, kv KV, tensors []Tensor) error {
return err return err
} }
padding := llm.padding(offset, 32) var alignment int64 = 32
if err := binary.Write(ws, llm.ByteOrder, bytes.Repeat([]byte{0}, int(padding-offset))); err != nil { padding := llm.padding(offset, alignment)
if err := binary.Write(ws, llm.ByteOrder, bytes.Repeat([]byte{0}, int(padding))); err != nil {
return err return err
} }
@ -638,8 +643,8 @@ func (llm *gguf) Encode(ws io.WriteSeeker, kv KV, tensors []Tensor) error {
return err return err
} }
padding := llm.padding(offset, 32) padding := llm.padding(offset, alignment)
if err := binary.Write(ws, llm.ByteOrder, bytes.Repeat([]byte{0}, int(padding-offset))); err != nil { if err := binary.Write(ws, llm.ByteOrder, bytes.Repeat([]byte{0}, int(padding))); err != nil {
return err return err
} }
} }
@ -648,5 +653,5 @@ func (llm *gguf) Encode(ws io.WriteSeeker, kv KV, tensors []Tensor) error {
} }
func (gguf) padding(offset, align int64) int64 { func (gguf) padding(offset, align int64) int64 {
return (offset + align - 1) / align * align return (align - offset%align) % align
} }