Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-27 06:39:25 +01:00
Code Issues Packages Projects Releases Wiki Activity

convert : fix Tensor type annotations

Browse Source
This commit is contained in:
Jared Van Bortel 2024-04-04 17:07:07 -04:00
parent 0d052cbe39
commit 6a9d3c0911
1 changed files with 8 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
convert.py
View File

@ -33,7 +33,7 @@ if 'NO_LOCAL_GGUF' not in os.environ:
import gguf
if TYPE_CHECKING:
from typing import TypeAlias
from typing_extensions import Self, TypeAlias
if hasattr(faulthandler, 'register') and hasattr(signal, 'SIGUSR1'):
faulthandler.register(signal.SIGUSR1)
@ -646,16 +646,17 @@ def permute(weights: NDArray, n_head: int, n_head_kv: int) -> NDArray:
class Tensor(ABC):
ndarray: NDArray
data_type: DataType
@abstractmethod
def astype(self, data_type: DataType) -> Tensor: ...
def astype(self, data_type: DataType) -> Self: ...
@abstractmethod
def permute(self, n_head: int, n_head_kv: int) -> Tensor: ...
def permute(self, n_head: int, n_head_kv: int) -> Self: ...
@abstractmethod
def permute_part(self, n_part: int, n_head: int, n_head_kv: int) -> UnquantizedTensor: ...
def permute_part(self, n_part: int, n_head: int, n_head_kv: int) -> Self: ...
@abstractmethod
def part(self, n_part: int) -> UnquantizedTensor: ...
def part(self, n_part: int) -> Self: ...
@abstractmethod
def to_ggml(self) -> GGMLCompatibleTensor: ...
@ -672,13 +673,13 @@ class UnquantizedTensor(Tensor):
self.ndarray = ndarray
self.data_type = NUMPY_TYPE_TO_DATA_TYPE[ndarray.dtype]
def astype(self, data_type: DataType) -> Tensor:
def astype(self, data_type: DataType) -> UnquantizedTensor:
dtype = data_type.dtype
if self.data_type == DT_BF16:
self.ndarray = bf16_to_fp32(self.ndarray)
return UnquantizedTensor(self.ndarray.astype(dtype))
def to_ggml(self) -> UnquantizedTensor:
def to_ggml(self) -> Self:
return self
def permute_part(self, n_part: int, n_head: int, n_head_kv: int) -> UnquantizedTensor: