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

Merge branch 'gguf-write-tokenization' into gguf

Browse Source
This commit is contained in:
M. Yusuf Sarıgöz 2023-07-29 00:36:35 +03:00
commit 9475cdb7a3
1 changed files with 32 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
gguf.py
View File

@ -15,7 +15,7 @@ import numpy as np
class GGMLQuantizationType(IntEnum):
F32 = 0
F16 = 1
QR_0 = 2
Q4_0 = 2
Q4_1 = 3
# Q4_2 = 4 # support has been removed
# Q4_3 = 5 # support has been removed
@ -75,7 +75,9 @@ class GGUFWriter:
return cls(f)
def write_key(self, key: str):
self.write_val(key, GGUFValueType.STRING)
encoded_key = key.encode("utf8")
self.fout.write(struct.pack("<I", len(encoded_key)))
self.fout.write(encoded_key)
def write_uint8(self, key: str, val: int):
self.write_key(key)
@ -158,28 +160,34 @@ class GGUFWriter:
return ((x + n - 1) // n) * n
def write_tensor_info(self, name: str, tensor: np.ndarray):
self.write_val(name, GGUFValueType.STRING)
self.write_key(name)
n_dims = len(tensor.shape)
self.write_val(n_dims, GGUFValueType.INT32)
self.fout.write(struct.pack("<i", n_dims))
for i in range(n_dims):
self.write_val(tensor.shape[n_dims - 1 - i], GGUFValueType.INT32)
self.fout.write(struct.pack("<i", tensor.shape[n_dims - 1 - i]))
assert tensor.dtype in (np.float32, np.float16), "Only F32 and F16 tensors are supported for now"
dtype = GGMLQuantizationType.F32 if tensor.dtype == np.float32 else GGMLQuantizationType.F16
self.write_val(dtype, GGUFValueType.INT32)
self.fout.write(struct.pack("<i", dtype))
self.fout.write(struct.pack("<Q", self.offset_tensor))
self.offset_tensor += GGUFWriter.ggml_pad(tensor.nbytes, constants.GGUF_DEFAULT_ALIGNMENT)
offset_data = GGUFWriter.ggml_pad(self.fout.tell(), constants.GGUF_DEFAULT_ALIGNMENT)
pad = offset_data - self.fout.tell()
self.fout.write(bytes([0] * pad))
self.flush()
self.tensors.append(tensor)
def write_tensors(self):
offset_data = GGUFWriter.ggml_pad(self.fout.tell(), constants.GGUF_DEFAULT_ALIGNMENT)
pad = offset_data - self.fout.tell()
print(f"pad: {pad}")
if pad != 0:
self.fout.write(bytes([0] * pad))
for tensor in self.tensors:
tensor.tofile(self.fout)
pad = GGUFWriter.ggml_pad(tensor.nbytes, constants.GGUF_DEFAULT_ALIGNMENT) - tensor.nbytes
print(f"pad: {pad}")
if pad != 0:
self.fout.write(bytes([0] * pad))
def flush(self):
@ -274,8 +282,8 @@ if __name__ == "__main__":
gguf_writer.write_architecture("llama")
gguf_writer.write_uint32("answer", 42) # Write a 32-bit integer
gguf_writer.write_float32("answer_in_float", 42.0) # Write a 32-bit float
tensor1 = np.random.random(size=(7, 10)).astype(np.float32)
tensor2 = np.random.random(size=(16, 12)).astype(np.float16)
tensor1 = np.ones((7, 8, 3), dtype=np.float32)
tensor2 = np.ones((7, 8, 3), dtype=np.float32)
gguf_writer.write_tensor_info("tensor1", tensor1)
gguf_writer.write_tensor_info("tensor2", tensor2)
gguf_writer.write_tensors()