root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-28 12:24:35 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

gguf : write tensors one by one and code reuse

Browse Source
This commit is contained in:
M. Yusuf Sarıgöz 2023-07-29 12:34:35 +03:00
parent aa99562d70
commit 93f7f7aef7
1 changed files with 6 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
gguf.py
View File

@ -61,7 +61,6 @@ class GGUFWriter:
def __init__(self, fout: IO): def __init__(self, fout: IO):
self.fout = fout self.fout = fout
self.offset_tensor = 0 self.offset_tensor = 0
self.tensors: List[np.ndarray] = []
def write_header(self, tensor_count: int, metadata_kv_count: int): def write_header(self, tensor_count: int, metadata_kv_count: int):
self.fout.write(struct.pack("<I", constants.GGUF_MAGIC)) self.fout.write(struct.pack("<I", constants.GGUF_MAGIC))
@ -75,9 +74,7 @@ class GGUFWriter:
return cls(f) return cls(f)
def write_key(self, key: str): def write_key(self, key: str):
encoded_key = key.encode("utf8") self.write_val(key, GGUFValueType.STRING, write_vtype=False)
self.fout.write(struct.pack("<I", len(encoded_key)))
self.fout.write(encoded_key)
def write_uint8(self, key: str, val: int): def write_uint8(self, key: str, val: int):
self.write_key(key) self.write_key(key)
@ -179,12 +176,11 @@ class GGUFWriter:
self.tensors.append(tensor) self.tensors.append(tensor)
def write_tensors(self): def write_tensor(self, tensor: np.ndarray):
pad = GGUFWriter.ggml_pad(self.fout.tell(), constants.GGUF_DEFAULT_ALIGNMENT) - self.fout.tell() pad = GGUFWriter.ggml_pad(self.fout.tell(), constants.GGUF_DEFAULT_ALIGNMENT) - self.fout.tell()
if pad != 0: if pad != 0:
self.fout.write(bytes([0] * pad)) self.fout.write(bytes([0] * pad))
for tensor in self.tensors:
tensor.tofile(self.fout) tensor.tofile(self.fout)
pad = GGUFWriter.ggml_pad(tensor.nbytes, constants.GGUF_DEFAULT_ALIGNMENT) - tensor.nbytes pad = GGUFWriter.ggml_pad(tensor.nbytes, constants.GGUF_DEFAULT_ALIGNMENT) - tensor.nbytes
if pad != 0: if pad != 0: