convert-hf : use an ABC for Model again

It seems Protocol can't be used as a statically type-checked ABC,
because its subclasses also can't be instantiated. (why did it seem to work?)

At least there's still a way to throw an error when forgetting to define
the `model_arch` property of any registered Model subclasses.

Author: compilade

convert-hf-to-gguf.py

@@ -8,10 +8,11 @@
 import os
 import re
 import sys
+from abc import ABC
 from enum import IntEnum
 from pathlib import Path
 from hashlib import sha256
-from typing import TYPE_CHECKING, Any, Callable, ContextManager, Iterable, Iterator, Protocol, Sequence, TypeVar, cast
+from typing import TYPE_CHECKING, Any, Callable, ContextManager, Iterable, Iterator, Sequence, TypeVar, cast
 
 import numpy as np
 import torch
@@ -40,7 +41,7 @@ class SentencePieceTokenTypes(IntEnum):
 AnyModel = TypeVar("AnyModel", bound="type[Model]")
 
 
-class Model(Protocol):
+class Model(ABC):
     _model_classes: dict[str, type[Model]] = {}
 
     dir_model: Path
@@ -57,6 +58,7 @@ class Model(Protocol):
     tensor_map: gguf.TensorNameMap
     tensors: dict[str, Tensor]
 
+    # subclasses should define this!
     model_arch: gguf.MODEL_ARCH
 
     def __init__(self, dir_model: Path, ftype: int, fname_out: Path, is_big_endian: bool, use_temp_file: bool):
@@ -163,12 +165,18 @@ def set_gguf_parameters(self):
         print(f"gguf: file type = {self.ftype}")
 
     def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
+        del bid  # unused
+
         return [(self.map_tensor_name(name), data_torch)]
 
     def extra_f32_tensors(self, name: str, new_name: str, bid: int | None, n_dims: int) -> bool:
+        del name, new_name, bid, n_dims  # unused
+
         return False
 
     def extra_f16_tensors(self, name: str, new_name: str, bid: int | None, n_dims: int) -> bool:
+        del name, new_name, bid, n_dims  # unused
+
         return False
 
     def write_tensors(self):
@@ -248,7 +256,7 @@ def get_model_part_names(dir_model: Path, suffix: str) -> list[str]:
         return part_names
 
     @staticmethod
-    def load_hparams(dir_model):
+    def load_hparams(dir_model: Path):
         with open(dir_model / "config.json", "r", encoding="utf-8") as f:
             return json.load(f)
 
@@ -258,12 +266,14 @@ def register(cls, *names: str) -> Callable[[AnyModel], AnyModel]:
 
         def func(modelcls: AnyModel) -> AnyModel:
             for name in names:
+                if "model_arch" not in modelcls.__dict__:
+                    raise TypeError(f"Missing property 'model_arch' for {modelcls.__name__!r}")
                 cls._model_classes[name] = modelcls
             return modelcls
         return func
 
     @classmethod
-    def from_model_architecture(cls, arch):
+    def from_model_architecture(cls, arch: str) -> type[Model]:
         try:
             return cls._model_classes[arch]
         except KeyError: