convert-*.py: lint check and add back in type templates

convert-hf-to-gguf.py

@@ -12,7 +12,7 @@
 from enum import IntEnum
 from pathlib import Path
 from hashlib import sha256
-from typing import TYPE_CHECKING, Any, Callable, ContextManager, Iterable, Iterator, Sequence, TypeVar, cast
+from typing import TYPE_CHECKING, Any, Callable, ContextManager, Iterable, Iterator, Sequence, TypeVar, cast, Optional
 from dataclasses import dataclass
 
 import math
@@ -148,18 +148,27 @@ def get_model_name(metadata, hyperparameters, dir_model, model_arch):
                 return gguf.MODEL_ARCH_NAMES[model_arch]
         self.model_name = get_model_name(self.metadata, self.hparams, self.dir_model, self.model_arch)
 
-        # Generate default filename based on model specification and available metadata
-        def extract_encoding_scheme(ftype):
-            # Extracts and converts the encoding scheme from the given file type name.
-            # e.g. 'gguf.LlamaFileType.ALL_F32' --> 'F32'
-            return ftype.name.partition("_")[2].upper()
+        # Extracts and converts the encoding scheme from the given file type name. e.g. 'gguf.LlamaFileType.ALL_F32' --> 'F32'
+        encodingScheme = self.ftype.name.partition("_")[2]
+
+        # Get Expert Count From Hyperparameters
         expert_count = self.hparams["num_local_experts"] if "num_local_experts" in self.hparams else None
-        self.fname_default = f"{gguf.naming_convention(self.model_name, self.metadata.version, expert_count, self.parameter_count(), extract_encoding_scheme(self.ftype))}"
+
+        # Generate default filename based on model specification and available metadata
+        self.fname_default = gguf.naming_convention(self.model_name, self.metadata.version, expert_count, self.parameter_count(), encodingScheme)
 
         # Filename Output
         if fname_out is not None:
             # custom defined filename and path was provided
-            self.fname_out = fname_out
+            def fill_templated_filename(filename: str, encodingScheme: str):
+                # Given a file name fill in any type templates e.g. 'some-model-name.{ftype}.gguf'
+                ftype_uppercase: str = encodingScheme.upper()
+                ftype_lowercase: str = encodingScheme.lower()
+                return filename.format(ftype_lowercase,
+                                       outtype=ftype_lowercase, ftype=ftype_lowercase,
+                                       OUTTYPE=ftype_uppercase, FTYPE=ftype_uppercase)
+
+            self.fname_out = fname_out.parent / fill_templated_filename(fname_out.name, encodingScheme)
         else:
             # output in the same directory as the model by default
             self.fname_out = dir_model.parent / self.fname_default

convert.py

@@ -1320,8 +1320,6 @@ def pick_output_type(model: LazyModel, output_type_str: str | None) -> GGMLFileT
 
 def model_parameter_count(model: LazyModel) -> int:
     # TODO: Ensure parameter count is accurate throughout various model type
-    #       May currently overestimate parameter count in Mamba model because
-    #       output weights is tied with token embeddings.
     total_model_parameters = 0
     for name, lazy_tensor in model.items():
         # Got A Tensor