sleepy/llama.cpp
1
0
Fork 0
Code Issues 9 Pull Requests Actions 103 Packages Projects Releases Wiki Activity

ci : switch from pyright to ty (#20826)

Browse Source 
* type fixes

* switch to ty

* tweak rules

* tweak more rules

* more tweaks

* final tweak

* use common import-not-found rule
This commit is contained in:
Sigbjørn Skjæret
2026-03-21 08:54:34 +01:00
committed by GitHub
parent cea560f483
commit 29b28a9824
20 changed files with 181 additions and 124 deletions
Download Patch File Download Diff File Expand all files Collapse all files
convert_hf_to_gguf.py
+13 -12
View File
@@ -31,10 +31,10 @@ import gguf
from gguf.vocab import MistralTokenizerType, MistralVocab
try:
from mistral_common.tokens.tokenizers.base import TokenizerVersion # pyright: ignore[reportMissingImports]
from mistral_common.tokens.tokenizers.multimodal import DATASET_MEAN as _MISTRAL_COMMON_DATASET_MEAN, DATASET_STD as _MISTRAL_COMMON_DATASET_STD # pyright: ignore[reportMissingImports]
from mistral_common.tokens.tokenizers.tekken import Tekkenizer # pyright: ignore[reportMissingImports]
from mistral_common.tokens.tokenizers.sentencepiece import ( # pyright: ignore[reportMissingImports]
from mistral_common.tokens.tokenizers.base import TokenizerVersion # type: ignore[import-not-found]
from mistral_common.tokens.tokenizers.multimodal import DATASET_MEAN as _MISTRAL_COMMON_DATASET_MEAN, DATASET_STD as _MISTRAL_COMMON_DATASET_STD # type: ignore[import-not-found]
from mistral_common.tokens.tokenizers.tekken import Tekkenizer # type: ignore[import-not-found]
from mistral_common.tokens.tokenizers.sentencepiece import ( # type: ignore[import-not-found]
SentencePieceTokenizer,
)
@@ -45,9 +45,9 @@ except ImportError:
_MISTRAL_COMMON_DATASET_STD = (0.26862954, 0.26130258, 0.27577711)
_mistral_common_installed = False
TokenizerVersion = None
Tekkenizer = None
SentencePieceTokenizer = None
TokenizerVersion: Any = None
Tekkenizer: Any = None
SentencePieceTokenizer: Any = None
_mistral_import_error_msg = (
"Mistral format requires `mistral-common` to be installed. Please run "
"`pip install mistral-common[image,audio]` to install it."
@@ -220,7 +220,7 @@ class ModelBase:
if weight_map is None or not isinstance(weight_map, dict):
raise ValueError(f"Can't load 'weight_map' from {index_name!r}")
tensor_names_from_index.update(weight_map.keys())
part_dict: dict[str, None] = dict.fromkeys(weight_map.values(), None)
part_dict: dict[str, None] = dict.fromkeys(weight_map.values(), None) # ty: ignore[invalid-assignment]
part_names = sorted(part_dict.keys())
else:
weight_map = {}
@@ -5882,7 +5882,7 @@ class InternLM2Model(TextModel):
logger.error(f'Error: Missing {tokenizer_path}')
sys.exit(1)
sentencepiece_model = model.ModelProto() # pyright: ignore[reportAttributeAccessIssue]
sentencepiece_model = model.ModelProto() # pyright: ignore[reportAttributeAccessIssue] # ty: ignore[unresolved-attribute]
sentencepiece_model.ParseFromString(open(tokenizer_path, "rb").read())
add_prefix = sentencepiece_model.normalizer_spec.add_dummy_prefix
@@ -6203,7 +6203,7 @@ class BertModel(TextModel):
vocab_size = max(self.hparams.get("vocab_size", 0), tokenizer.vocab_size)
else:
sentencepiece_model = model.ModelProto() # pyright: ignore[reportAttributeAccessIssue]
sentencepiece_model = model.ModelProto() # pyright: ignore[reportAttributeAccessIssue] # ty: ignore[unresolved-attribute]
sentencepiece_model.ParseFromString(open(tokenizer_path, "rb").read())
assert sentencepiece_model.trainer_spec.model_type == 1 # UNIGRAM
@@ -8880,7 +8880,7 @@ class T5Model(TextModel):
if not tokenizer_path.is_file():
raise FileNotFoundError(f"File not found: {tokenizer_path}")
sentencepiece_model = model.ModelProto() # pyright: ignore[reportAttributeAccessIssue]
sentencepiece_model = model.ModelProto() # pyright: ignore[reportAttributeAccessIssue] # ty: ignore[unresolved-attribute]
sentencepiece_model.ParseFromString(open(tokenizer_path, "rb").read())
# some models like Pile-T5 family use BPE tokenizer instead of Unigram
@@ -9017,7 +9017,7 @@ class T5EncoderModel(TextModel):
if not tokenizer_path.is_file():
raise FileNotFoundError(f"File not found: {tokenizer_path}")
sentencepiece_model = model.ModelProto() # pyright: ignore[reportAttributeAccessIssue]
sentencepiece_model = model.ModelProto() # pyright: ignore[reportAttributeAccessIssue] # ty: ignore[unresolved-attribute]
sentencepiece_model.ParseFromString(open(tokenizer_path, "rb").read())
# some models like Pile-T5 family use BPE tokenizer instead of Unigram
@@ -12279,6 +12279,7 @@ class LazyTorchTensor(gguf.LazyBase):
kwargs = {}
if func is torch.Tensor.numpy:
assert len(args)
return args[0].numpy()
return cls._wrap_fn(func)(*args, **kwargs)