tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-26 19:55:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

gguf : enforce that tensor names are unique (#6905)

Browse Source 
* not allow adding duplicated tensor name

* no duplicated tensor while reading gguf

* typo

* throw exception inside llama_model_loader

Co-authored-by: slaren <slarengh@gmail.com>

---------

Co-authored-by: slaren <slarengh@gmail.com>
This commit is contained in:
Xuan Son Nguyen
2024-04-28 17:36:18 +02:00
committed by GitHub
parent ce023f6f2f
commit 7bb36ccf91
4 changed files with 32 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
gguf-py/gguf/gguf_reader.py
View File

@ -234,8 +234,14 @@ class GGUFReader:
def _build_tensors(self, start_offs: int, fields: list[ReaderField]) -> None:
tensors = []
tensor_names = set() # keep track of name to prevent duplicated tensors
for field in fields:
_name_len, name_data, _n_dims, dims, raw_dtype, offset_tensor = field.parts
# check if there's any tensor having same name already in the list
tensor_name = str(bytes(name_data), encoding = 'utf-8')
if tensor_name in tensor_names:
raise ValueError(f'Found duplicated tensor with name {tensor_name}')
tensor_names.add(tensor_name)
ggml_type = GGMLQuantizationType(raw_dtype[0])
n_elems = np.prod(dims)
block_size, type_size = GGML_QUANT_SIZES[ggml_type]
@ -267,7 +273,7 @@ class GGUFReader:
item_count = n_bytes
item_type = np.uint8
tensors.append(ReaderTensor(
name = str(bytes(name_data), encoding = 'utf-8'),
name = tensor_name,
tensor_type = ggml_type,
shape = dims,
n_elements = n_elems,