Skip to content

fix(gguf_parser): fix memoryerror exception when loading non-native models #1452

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
May 30, 2025
Merged

fix(gguf_parser): fix memoryerror exception when loading non-native models #1452

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
26 changes: 17 additions & 9 deletions ramalama/gguf_parser.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -104,21 +104,28 @@ class ParseError(Exception):


class GGUFInfoParser:
@staticmethod
def is_model_gguf(model_path: str) -> bool:
try:
with open(model_path, "rb") as model_file:
magic_number = GGUFInfoParser.read_string(model_file, 4)
magic_number = GGUFInfoParser.read_string(model_file, GGUFEndian.LITTLE, 4)
return magic_number == GGUFModelInfo.MAGIC_NUMBER
except Exception as ex:
console.warning(f"Failed to read model '{model_path}': {ex}")
return False

@staticmethod
def read_string(model: io.BufferedReader, length: int = -1) -> str:
def read_string(
model: io.BufferedReader, model_endianness: GGUFEndian = GGUFEndian.LITTLE, length: int = -1
) -> str:
if length == -1:
type_string = GGUF_VALUE_TYPE_FORMAT[GGUFValueType.UINT64]
length = struct.unpack(type_string, model.read(struct.calcsize(type_string)))[0]
return model.read(length).decode("utf-8")
length = GGUFInfoParser.read_number(model, GGUFValueType.UINT64, model_endianness)

raw = model.read(length)
if len(raw) < length:
raise ParseError(f"Unexpected EOF: wanted {length} bytes, got {len(raw)}")

return raw.decode("utf-8")

@staticmethod
def read_number(model: io.BufferedReader, value_type: GGUFValueType, model_endianness: GGUFEndian) -> float:
Expand Down Expand Up @@ -151,7 +158,7 @@ def read_value(model: io.BufferedReader, value_type: GGUFValueType, model_endian
elif value_type == GGUFValueType.BOOL:
value = GGUFInfoParser.read_bool(model, model_endianness)
elif value_type == GGUFValueType.STRING:
value = GGUFInfoParser.read_string(model)
value = GGUFInfoParser.read_string(model, model_endianness)
elif value_type == GGUFValueType.ARRAY:
array_type = GGUFInfoParser.read_value_type(model, model_endianness)
array_length = GGUFInfoParser.read_number(model, GGUFValueType.UINT64, model_endianness)
Expand All @@ -161,13 +168,14 @@ def read_value(model: io.BufferedReader, value_type: GGUFValueType, model_endian
return value
raise ParseError(f"Unknown type '{value_type}'")

@staticmethod
def parse(model_name: str, model_registry: str, model_path: str) -> GGUFModelInfo:
# Pin model endianness to Little Endian by default.
# Models downloaded via HuggingFace are majority Little Endian.
model_endianness = GGUFEndian.LITTLE

with open(model_path, "rb") as model:
magic_number = GGUFInfoParser.read_string(model, 4)
magic_number = GGUFInfoParser.read_string(model, model_endianness, 4)
if magic_number != GGUFModelInfo.MAGIC_NUMBER:
raise ParseError(f"Invalid GGUF magic number '{magic_number}'")

Expand All @@ -185,13 +193,13 @@ def parse(model_name: str, model_registry: str, model_path: str) -> GGUFModelInf

metadata = {}
for _ in range(metadata_kv_count):
key = GGUFInfoParser.read_string(model)
key = GGUFInfoParser.read_string(model, model_endianness)
value_type = GGUFInfoParser.read_value_type(model, model_endianness)
metadata[key] = GGUFInfoParser.read_value(model, value_type, model_endianness)

tensors: list[Tensor] = []
for _ in range(tensor_count):
name = GGUFInfoParser.read_string(model)
name = GGUFInfoParser.read_string(model, model_endianness)
n_dimensions = GGUFInfoParser.read_number(model, GGUFValueType.UINT32, model_endianness)
dimensions: list[int] = []
for _ in range(n_dimensions):
Expand Down
Loading