Add typo checker (#846)

* Add typo checker

Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

Author: XuehaoSun

.pre-commit-config.yaml

@@ -51,6 +51,11 @@ repos:
         additional_dependencies:
           - tomli
 
+  - repo: https://github.com/crate-ci/typos
+    rev: v1.36.2
+    hooks:
+      - id: typos
+
   - repo: https://github.com/pycqa/isort
     rev: 6.0.1
     hooks:

README.md

@@ -6,7 +6,7 @@ AutoRound
 <h3> Advanced Quantization Algorithm for LLMs</h3>
 
 [![python](https://img.shields.io/badge/python-3.10%2B-blue)](https://github.com/intel/auto-round)
-[![version](https://img.shields.io/badge/release-0.7.0-green)](https://github.com/intel/auto-round)
+[![version](https://img.shields.io/badge/release-0.7.1-green)](https://github.com/intel/auto-round)
 [![license](https://img.shields.io/badge/license-Apache%202-9C27B0)](https://github.com/intel/auto-round/blob/main/LICENSE)
 <a href="https://huggingface.co/Intel">
 <img alt="Model Checkpoints" src="https://img.shields.io/badge/%F0%9F%A4%97%20HF-Models-F57C00">

auto_round/compressors/base.py

@@ -94,7 +94,7 @@
     set_module,
     to_device,
     to_dtype,
-    unsupport_meta_device,
+    unsupported_meta_device,
 )
 from auto_round.wrapper import WrapperLinear, WrapperMultiblock, unwrapper_block, unwrapper_layer, wrapper_block
 
@@ -260,7 +260,7 @@ def __init__(
         elif tokenizer is None and iters > 0:
             raise ValueError("A tokenizer must be set for non-str model input")
         self.low_cpu_mem_usage = bool(low_cpu_mem_usage)
-        if unsupport_meta_device(model):
+        if unsupported_meta_device(model):
             raise RuntimeError(
                 "AutoRound does not support parameters on meta device. "
                 "Please use more GPUs by setting `--device 0,1,2,3` or just place the model on CPU."
@@ -345,7 +345,7 @@ def __init__(
         elif tokenizer is None and iters > 0:
             raise ValueError("A tokenizer must be set for non-str model input")
         self.low_cpu_mem_usage = bool(low_cpu_mem_usage)
-        if unsupport_meta_device(model):
+        if unsupported_meta_device(model):
             raise RuntimeError(
                 "AutoRound does not support parameters on meta device. "
                 "Please use more GPUs by setting `--device_map 0,1,2,3` or just place the model on CPU."
@@ -624,20 +624,20 @@ def _set_auto_device_map_in_block(self, block: torch.nn.Module, input_ids: list[
         device_0_memory = get_device_memory(
             self.device_list[0] if hasattr(self, "device_list") and self.device_list else 0
         )
-        block_memory, input_ouput_memory = estimate_tuning_block_mem(block, input_ids)
+        block_memory, input_output_memory = estimate_tuning_block_mem(block, input_ids)
         if self.low_gpu_mem_usage:
-            input_ouput_memory = 0
+            input_output_memory = 0
 
         mem_per_param_scale = 13 if self.mem_per_param_scale is None else self.mem_per_param_scale
         if self.iters == 0:
             mem_per_param_scale = 1  # for rtn
 
-        if (block_memory * mem_per_param_scale + input_ouput_memory) < device_0_memory:
+        if (block_memory * mem_per_param_scale + input_output_memory) < device_0_memory:
             return  # fit in one GPU
 
         device_map = {}
         device_memory = {device: get_device_memory(int(device.split(":")[1])) for device in cuda_devices}
-        device_memory[device_0] = device_0_memory - input_ouput_memory
+        device_memory[device_0] = device_0_memory - input_output_memory
 
         device_idx = 0
         # First, fill device 0 to its maximum capacity, then distribute the remaining layers evenly across other devices
@@ -864,7 +864,7 @@ def remove_duplicates(lst):
                         format = "auto_round:auto_awq"
                 elif is_nv_fp(self.data_type) or is_mx_fp(self.data_type):
                     format = f"auto_round:{self.data_type}"
-                elif is_static_wfp8afp8(self):  # staic wfp8afp8
+                elif is_static_wfp8afp8(self):  # static wfp8afp8
                     format = f"auto_round:{AutoRoundFormat.FP8_STATIC.value}"
                 elif self.data_type == "fp" and self.bits == 8 and self.act_bits >= 16:  # woq fp8
                     format = f"auto_round:{AutoRoundFormat.FP8.value}"

auto_round/export/export_to_gguf/convert_hf_to_gguf.py

@@ -7458,7 +7458,7 @@ def set_gguf_parameters(self):
         layer_norm_eps = hparams["layer_norm_epsilon"]
         intermediate_size = hparams["intermediate_size"] if "intermediate_size" in hparams else 4 * embed_dim
         num_layers = hparams["num_layers"]
-        # ignore for now as EXAONE-3.0-7.8B-Instruct attentino_dropout is 0.0
+        # ignore for now as EXAONE-3.0-7.8B-Instruct attention_dropout is 0.0
         # attention_dropout_rate = hparams["attention_dropout"]
         # ignore for now as EXAONE-3.0-7.8B-Instruct embed_dropout is 0.0
         # embed_dropout_rate = hparams["embed_dropout"]
@@ -7707,7 +7707,7 @@ def __init__(self, *args, **kwargs):
     def get_attn_layers(self):
         # Explicit list of layer type names
         if layer_types := self.hparams.get("layer_types"):
-            return [i for i, typ in enumerate(layer_types) if typ == "attention"]
+            return [i for i, layer_type in enumerate(layer_types) if layer_type == "attention"]
 
         # Layer types indicated by index or period
         attn_layers = self.hparams.get("attn_layer_indices", [])

auto_round/export/export_to_gguf/packing.py

@@ -751,11 +751,11 @@ def q6_k_quant_block(blocks: np.array, scale=None, d_scale=None, original=False,
 
     tmp_L = all_L.reshape(nb, 4, 64) & 0xF
     output_ql = (tmp_L[:, ::2] | (tmp_L[:, 1::2] << 4)).reshape(nb, QK_K // 2).cpu().numpy().astype(np.uint8)
-    ouptut_qh = (all_L >> 4).reshape(nb, 2, 4, 32) << torch.tensor([0, 2, 4, 6], device=all_L.device).reshape(
+    output_qh = (all_L >> 4).reshape(nb, 2, 4, 32) << torch.tensor([0, 2, 4, 6], device=all_L.device).reshape(
         1, 1, 4, 1
     )
     output_qh = (
-        np.bitwise_or.reduce(ouptut_qh.cpu().numpy(), axis=2, dtype=np.uint8)  # pylint: disable=E1121
+        np.bitwise_or.reduce(output_qh.cpu().numpy(), axis=2, dtype=np.uint8)  # pylint: disable=E1121
         .reshape(nb, QK_K // 4)
         .astype(np.uint8)
     )  # pylint: disable=E1121

auto_round/inference/auto_quantizer.py

@@ -270,8 +270,8 @@ def post_init(self):
             raise ValueError("group_size must be greater than 0 or equal to -1")
 
     def get_loading_attributes(self):
-        loading_attibutes_dict = {"backend": self.backend}
-        return loading_attibutes_dict
+        loading_attributes_dict = {"backend": self.backend}
+        return loading_attributes_dict
 
     def to_dict(self):
         config_dict = super().to_dict()

auto_round/inference/backend.py

@@ -788,7 +788,7 @@ def find_backend(backend: str, orig_backend: str = None):
         target_info = BackendInfos[key]
         if (
             target_info.packing_format == orig_info.packing_format
-            or orig_info.packing_format in target_info.convertable_format
+            or orig_info.packing_format in target_info.convertible_format
         ):
             return key
 

auto_round/low_cpu_mem/utils.py

@@ -423,16 +423,16 @@ def _layer_wise_to(module, name, device_or_dtype):
             module.get_bias = partial(_get_value, name, "bias")
         module.update = partial(_update, name, module)
 
-    def _repalce_to(module, name):
+    def _replace_to(module, name):
         if len(module._modules) > 0:
             for n, m in module.named_children():
                 if len(name) > 0:
                     n = name + "." + n
-                _repalce_to(m, n)
+                _replace_to(m, n)
         module.ori_to = module.to
         module.to = partial(_layer_wise_to, module, name)
 
-    _repalce_to(empty_model, "")
+    _replace_to(empty_model, "")
 
 
 def load_model_with_hooks(

auto_round/testing_utils.py

@@ -47,7 +47,7 @@ def is_itrex_available():
     return importlib.util.find_spec("intel_extension_for_transformers") is not None
 
 
-def is_flash_attn_avaliable():
+def is_flash_attn_available():
     return importlib.util.find_spec("flash_attn") is not None
 
 
@@ -203,7 +203,7 @@ def require_vlm_env(test_case):
 
     env_check = True
     # pip install flash-attn --no-build-isolation
-    env_check &= is_flash_attn_avaliable()
+    env_check &= is_flash_attn_available()
 
     # pip install git+https://github.com/haotian-liu/LLaVA.git@v1.2.2
     env_check &= importlib.util.find_spec("llava") is not None

auto_round/utils.py

@@ -220,7 +220,7 @@ def get_scale_shape(weight, group_size):
     return shape
 
 
-def unsupport_meta_device(model):
+def unsupported_meta_device(model):
     """Checks if the model is a valid model for auto_round.
 
     Args:
@@ -810,8 +810,8 @@ def is_autoround_exllamav2_available():
 def get_library_version(library_name):
     from packaging.version import Version
 
-    python_vesion = Version(sys.version.split()[0])
-    if python_vesion < Version("3.8"):
+    python_version = Version(sys.version.split()[0])
+    if python_version < Version("3.8"):
         import warnings
 
         warnings.filterwarnings("ignore", category=DeprecationWarning)
@@ -1290,7 +1290,7 @@ def _gguf_args_check(args_or_ar, formats: list[str] = None, model_type=ModelType
 
     pattern = re.compile(r"q\d_k")
     pre_dq_format = ""
-    unsupport_list, reset_list = [], []
+    unsupported_list, reset_list = [], []
     for format in GGUF_CONFIG:
         if format in formats:
             if format == "q6_k_s":
@@ -1303,7 +1303,7 @@ def _gguf_args_check(args_or_ar, formats: list[str] = None, model_type=ModelType
                 else:
                     pre_dq_format = format
 
-            unsupport_list, reset_list = [], []
+            unsupported_list, reset_list = [], []
             gguf_config = GGUF_CONFIG[format]
             for k, v in gguf_config.items():
                 if not hasattr(args_or_ar, k):
@@ -1315,12 +1315,12 @@ def _gguf_args_check(args_or_ar, formats: list[str] = None, model_type=ModelType
                     k = "asym"
                     v = not v
                 if getattr(args_or_ar, k) != v:
-                    unsupport_list.append(f"{k}={getattr(args_or_ar, k)}")
+                    unsupported_list.append(f"{k}={getattr(args_or_ar, k)}")
                     reset_list.append(f"{k}={v}")
                     setattr(args_or_ar, k, v)
-            if len(unsupport_list) > 0:
+            if len(unsupported_list) > 0:
                 logger.info(
-                    f"format {format} does not support for {', '.join(unsupport_list)},"
+                    f"format {format} does not support for {', '.join(unsupported_list)},"
                     f" reset to {', '.join(reset_list)}."
                 )
     # Removed obsolete commented-out block for improved readability and maintainability.
@@ -2415,7 +2415,7 @@ def module_match_name_list(module, name_list):
     elif module_match_name_list(module, ["DBRXMoeSparseMoeBlock"]):
         return ["w1_linear", "w2_linear", "v1_linear"]
     else:
-        # assuing w1, w2, w3 by default
+        # assuming w1, w2, w3 by default
         return ["w1", "w2", "w3"]