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

1. check in ggml.c if endianess is not match

Browse source 
2. update GGUF version
3. change get_pack_prefix to property
4. update information log
This commit is contained in:
chenqiny 2023-10-13 00:23:16 +08:00
parent 51e9d39117
commit 7fc0250d15
5 changed files with 39 additions and 29 deletions
Download patch file Download diff file Expand all files Collapse all files

1
convert-baichuan-hf-to-gguf.py
View file

@ -90,6 +90,7 @@ if not dir_model.is_dir():
endianess = gguf.GGUFEndian.LITTLE
if args.bigendian:
endianess = gguf.GGUFEndian.BIG
endianess_str = "Big Endian" if args.bigendian else "Little Endian"
print(f"gguf: Conversion Endianess {endianess}")
# possible tensor data types
# ftype == 0 -> float32

8
ggml.c
View file

@ -20916,7 +20916,13 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
gguf_fread_el(file, &magic, sizeof(magic), &offset);
if (magic != GGUF_MAGIC) {
fprintf(stderr, "%s: invalid magic number %08x\n", __func__, magic);
if (magic == GGUF_WRONG_ENIAN_MAGIC)
{
fprintf(stderr, "Endianess of the GGUF file and platform do not match.%s: invalid magic number %08x.\n", __func__, magic);
}
else {
fprintf(stderr, "%s: invalid magic number %08x\n", __func__, magic);
}
fclose(file);
return NULL;
}

3
ggml.h
View file

@ -232,7 +232,8 @@
#define GGML_EXIT_ABORTED 1
#define GGUF_MAGIC 0x46554747 // "GGUF"
#define GGUF_VERSION 2
#define GGUF_WRONG_ENIAN_MAGIC 0x47475546
#define GGUF_VERSION 3
#define GGUF_DEFAULT_ALIGNMENT 32

54
gguf-py/gguf/gguf.py
View file

@ -19,7 +19,7 @@ import numpy as np
#
GGUF_MAGIC = 0x46554747
GGUF_VERSION = 2
GGUF_VERSION = 3
GGUF_DEFAULT_ALIGNMENT = 32
@ -621,7 +621,8 @@ class GGUFWriter:
temp_file: tempfile.SpooledTemporaryFile[bytes] | None = None
tensors: list[tuple[np.ndarray[Any, Any], int]]
def get_pack_prefix(self):
@property
def pack_prefix(self):
if self.endianess==GGUFEndian.LITTLE:
return "<"
else:
@ -632,28 +633,29 @@ class GGUFWriter:
self.arch = arch
self.endianess = endianess
self._simple_value_packing = {
GGUFValueType.UINT8: f"{self.get_pack_prefix()}B",
GGUFValueType.INT8: f"{self.get_pack_prefix()}b",
GGUFValueType.UINT16: f"{self.get_pack_prefix()}H",
GGUFValueType.INT16: f"{self.get_pack_prefix()}h",
GGUFValueType.UINT32: f"{self.get_pack_prefix()}I",
GGUFValueType.INT32: f"{self.get_pack_prefix()}i",
GGUFValueType.FLOAT32: f"{self.get_pack_prefix()}f",
GGUFValueType.UINT64: f"{self.get_pack_prefix()}Q",
GGUFValueType.INT64: f"{self.get_pack_prefix()}q",
GGUFValueType.FLOAT64: f"{self.get_pack_prefix()}d",
GGUFValueType.UINT8: f"{self.pack_prefix}B",
GGUFValueType.INT8: f"{self.pack_prefix}b",
GGUFValueType.UINT16: f"{self.pack_prefix}H",
GGUFValueType.INT16: f"{self.pack_prefix}h",
GGUFValueType.UINT32: f"{self.pack_prefix}I",
GGUFValueType.INT32: f"{self.pack_prefix}i",
GGUFValueType.FLOAT32: f"{self.pack_prefix}f",
GGUFValueType.UINT64: f"{self.pack_prefix}Q",
GGUFValueType.INT64: f"{self.pack_prefix}q",
GGUFValueType.FLOAT64: f"{self.pack_prefix}d",
GGUFValueType.BOOL: "?" ,
}
self.add_architecture()
self.use_temp_file = use_temp_file
self.tensors = []
print(f"This gguf file is for {self.endianess} only")
endianess_str = "Big Endian" if self.endianess == GGUFEndian.BIG else "Little Endian"
print(f"This gguf file is for {endianess_str} only")
def write_header_to_file(self):
self.fout.write(struct.pack(f"{self.get_pack_prefix()}I", GGUF_MAGIC))
self.fout.write(struct.pack(f"{self.get_pack_prefix()}I", GGUF_VERSION))
self.fout.write(struct.pack(f"{self.get_pack_prefix()}Q", self.ti_data_count))
self.fout.write(struct.pack(f"{self.get_pack_prefix()}Q", self.kv_data_count))
self.fout.write(struct.pack(f"{self.pack_prefix}I", GGUF_MAGIC))
self.fout.write(struct.pack(f"{self.pack_prefix}I", GGUF_VERSION))
self.fout.write(struct.pack(f"{self.pack_prefix}Q", self.ti_data_count))
self.fout.write(struct.pack(f"{self.pack_prefix}Q", self.kv_data_count))
self.flush()
# print("tensors " + str(self.ti_data_count) + " kv " + str(self.kv_data_count))
@ -730,7 +732,7 @@ class GGUFWriter:
vtype = GGUFValueType.get_type(val)
if add_vtype:
self.kv_data += struct.pack(f"{self.get_pack_prefix()}I", vtype)
self.kv_data += struct.pack(f"{self.pack_prefix}I", vtype)
self.kv_data_count += 1
pack_fmt = self._simple_value_packing.get(vtype)
@ -738,14 +740,14 @@ class GGUFWriter:
self.kv_data += struct.pack(pack_fmt, val)
elif vtype == GGUFValueType.STRING:
encoded_val = val.encode("utf8") if isinstance(val, str) else val
self.kv_data += struct.pack(f"{self.get_pack_prefix()}Q", len(encoded_val))
self.kv_data += struct.pack(f"{self.pack_prefix}Q", len(encoded_val))
self.kv_data += encoded_val
elif vtype == GGUFValueType.ARRAY and isinstance(val, Sequence) and len(val) > 0:
ltype = GGUFValueType.get_type(val[0])
if not all(GGUFValueType.get_type(i) is ltype for i in val[1:]):
raise ValueError("All items in a GGUF array should be of the same type")
self.kv_data += struct.pack(f"{self.get_pack_prefix()}I", ltype)
self.kv_data += struct.pack(f"{self.get_pack_prefix()}Q", len(val))
self.kv_data += struct.pack(f"{self.pack_prefix}I", ltype)
self.kv_data += struct.pack(f"{self.pack_prefix}Q", len(val))
for item in val:
self.add_val(item, add_vtype=False)
else:
@ -759,18 +761,18 @@ class GGUFWriter:
assert raw_dtype is not None or tensor_dtype in (np.float32, np.float16), "Only F32 and F16 tensors are supported for now"
encoded_name = name.encode("utf8")
self.ti_data += struct.pack(f"{self.get_pack_prefix()}Q", len(encoded_name))
self.ti_data += struct.pack(f"{self.pack_prefix}Q", len(encoded_name))
self.ti_data += encoded_name
n_dims = len(tensor_shape)
self.ti_data += struct.pack(f"{self.get_pack_prefix()}I", n_dims)
self.ti_data += struct.pack(f"{self.pack_prefix}I", n_dims)
for i in range(n_dims):
self.ti_data += struct.pack(f"{self.get_pack_prefix()}Q", tensor_shape[n_dims - 1 - i])
self.ti_data += struct.pack(f"{self.pack_prefix}Q", tensor_shape[n_dims - 1 - i])
if raw_dtype is None:
dtype = GGMLQuantizationType.F32 if tensor_dtype == np.float32 else GGMLQuantizationType.F16
else:
dtype = raw_dtype
self.ti_data += struct.pack(f"{self.get_pack_prefix()}I", dtype)
self.ti_data += struct.pack(f"{self.get_pack_prefix()}Q", self.offset_tensor)
self.ti_data += struct.pack(f"{self.pack_prefix}I", dtype)
self.ti_data += struct.pack(f"{self.pack_prefix}Q", self.offset_tensor)
self.offset_tensor += GGUFWriter.ggml_pad(tensor_nbytes, self.data_alignment)
self.ti_data_count += 1

2
gguf-py/pyproject.toml
View file

@ -1,6 +1,6 @@
[tool.poetry]
name = "gguf"
version = "0.4.4"
version = "0.4.5"
description = "Write ML models in GGUF for GGML"
authors = ["GGML <ggml@ggml.ai>"]
packages = [