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

convert_hf_to_gguf.py: Remove code that is already in fill_templated_filename() and GGUFWriter()

Browse source
This commit is contained in:
brian khuu 2024-07-07 20:00:26 +10:00
parent 2f23927d37
commit 4dc8ddd35a
1 changed files with 4 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

12
convert_hf_to_gguf.py
View file

@ -90,6 +90,7 @@ class Model:
self.tensor_names = None self.tensor_names = None
self.metadata = metadata self.metadata = metadata
# Apply heuristics to figure out typical tensor encoding based on first layer tensor encoding type
if self.ftype == gguf.LlamaFileType.GUESSED: if self.ftype == gguf.LlamaFileType.GUESSED:
# NOTE: can't use field "torch_dtype" in config.json, because some finetunes lie. # NOTE: can't use field "torch_dtype" in config.json, because some finetunes lie.
_, first_tensor = next(self.get_tensors()) _, first_tensor = next(self.get_tensors())
@ -100,13 +101,6 @@ class Model:
logger.info(f"choosing --outtype bf16 from first tensor type ({first_tensor.dtype})") logger.info(f"choosing --outtype bf16 from first tensor type ({first_tensor.dtype})")
self.ftype = gguf.LlamaFileType.MOSTLY_BF16 self.ftype = gguf.LlamaFileType.MOSTLY_BF16
ftype_up: str = self.ftype.name.partition("_")[2].upper()
ftype_lw: str = ftype_up.lower()
# allow templating the file name with the output ftype, useful with the "auto" ftype
self.fname_out = fname_out.parent / fname_out.name.format(ftype_lw, outtype=ftype_lw, ftype=ftype_lw, OUTTYPE=ftype_up, FTYPE=ftype_up)
self.gguf_writer = gguf.GGUFWriter(path=None, arch=gguf.MODEL_ARCH_NAMES[self.model_arch], endianess=self.endianess, use_temp_file=self.use_temp_file,
split_max_tensors=split_max_tensors, split_max_size=split_max_size, dry_run=dry_run, small_first_shard=small_first_shard)
# Fallback to model architecture name if metadata name is still missing # Fallback to model architecture name if metadata name is still missing
if self.metadata.name is None: if self.metadata.name is None:
self.metadata.name = gguf.MODEL_ARCH_NAMES[self.model_arch] self.metadata.name = gguf.MODEL_ARCH_NAMES[self.model_arch]
@ -126,13 +120,15 @@ class Model:
# Filename Output # Filename Output
if fname_out is not None: if fname_out is not None:
# custom defined filename and path was provided # custom defined filename and path was provided
# allow templating the file name with the output ftype, useful with the "auto" ftype
self.fname_out = fname_out.parent / gguf.fill_templated_filename(fname_out.name, output_type) self.fname_out = fname_out.parent / gguf.fill_templated_filename(fname_out.name, output_type)
else: else:
# output in the same directory as the model by default # output in the same directory as the model by default
self.fname_out = dir_model / f"{self.fname_default}.gguf" self.fname_out = dir_model / f"{self.fname_default}.gguf"
# Configure GGUF Writer # Configure GGUF Writer
self.gguf_writer = gguf.GGUFWriter(self.fname_out, gguf.MODEL_ARCH_NAMES[self.model_arch], endianess=self.endianess, use_temp_file=self.use_temp_file) self.gguf_writer = gguf.GGUFWriter(path=self.fname_out, arch=gguf.MODEL_ARCH_NAMES[self.model_arch], endianess=self.endianess, use_temp_file=self.use_temp_file,
split_max_tensors=split_max_tensors, split_max_size=split_max_size, dry_run=dry_run, small_first_shard=small_first_shard)
@classmethod @classmethod
def __init_subclass__(cls): def __init_subclass__(cls):