vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

clean code 2

Browse source
This commit is contained in:
root 2024-06-09 21:15:02 +08:00
parent 1c5a8b7fec
commit 3a0f8b0697
5 changed files with 100 additions and 51 deletions
Download patch file Download diff file Expand all files Collapse all files

3
gguf-py/gguf/constants.py
View file

@ -925,7 +925,6 @@ class GGMLQuantizationType(IntEnum):
F64 = 28
IQ1_M = 29
BF16 = 30
I2 = 31
# TODO: add GGMLFileType from ggml_ftype in ggml.h
@ -967,7 +966,6 @@ class LlamaFileType(IntEnum):
MOSTLY_IQ4_XS = 30 # except 1d tensors
MOSTLY_IQ1_M = 31 # except 1d tensors
MOSTLY_BF16 = 32 # except 1d tensors
MOSTLY_I2 = 33 # except 1d tensors
GUESSED = 1024 # not specified in the model file
@ -1034,7 +1032,6 @@ GGML_QUANT_SIZES: dict[GGMLQuantizationType, tuple[int, int]] = {
GGMLQuantizationType.IQ3_S: (256, 2 + QK_K // 4 + QK_K // 8 + QK_K // 32 + 4),
GGMLQuantizationType.IQ2_S: (256, 2 + QK_K // 4 + QK_K // 16),
GGMLQuantizationType.IQ4_XS: (256, 2 + 2 + QK_K // 2 + QK_K // 64),
GGMLQuantizationType.I2: (1, 1),
GGMLQuantizationType.I8: (1, 1),
GGMLQuantizationType.I16: (1, 2),
GGMLQuantizationType.I32: (1, 4),

13
gguf-py/gguf/gguf_writer.py
View file

@ -225,10 +225,8 @@ class GGUFWriter:
dtype = GGMLQuantizationType.I32
elif tensor_dtype == np.int64:
dtype = GGMLQuantizationType.I64
elif tensor_dtype == np.uint8:
dtype = GGMLQuantizationType.I2
else:
raise ValueError("Only F16, F32, F64, I8, I16, I32, I64, I2 tensors are supported for now")
raise ValueError("Only F16, F32, F64, I8, I16, I32, I64 tensors are supported for now")
else:
dtype = raw_dtype
if tensor_dtype == np.uint8:
@ -239,10 +237,7 @@ class GGUFWriter:
self.ti_data += self._pack("Q", tensor_shape[n_dims - 1 - i])
self.ti_data += self._pack("I", dtype)
self.ti_data += self._pack("Q", self.offset_tensor)
if dtype == GGMLQuantizationType.I2:
self.offset_tensor += GGUFWriter.ggml_pad(tensor_nbytes, self.data_alignment) + self.data_alignment
else:
self.offset_tensor += GGUFWriter.ggml_pad(tensor_nbytes, self.data_alignment)
self.offset_tensor += GGUFWriter.ggml_pad(tensor_nbytes, self.data_alignment)
self.ti_data_count += 1
def add_tensor(
@ -257,9 +252,7 @@ class GGUFWriter:
self.temp_file = fp
shape: Sequence[int] = raw_shape if raw_shape is not None else tensor.shape
if (raw_dtype != GGMLQuantizationType.F32 or not name.endswith("scale")):
self.add_tensor_info(name, shape, tensor.dtype, tensor.nbytes, raw_dtype = raw_dtype)
self.add_tensor_info(name, shape, tensor.dtype, tensor.nbytes, raw_dtype = raw_dtype)
if self.temp_file is None:
self.tensors.append(tensor)