Fix lints (#456)

Author: mergennachin

.ci/scripts/gather_test_models.py

@@ -19,7 +19,6 @@
     "mistralai/Mistral-7B-v0.1": "https://huggingface.co/mistralai/Mistral-7B-v0.1/resolve/main/config.json,https://huggingface.co/mistralai/Mistral-7B-v0.1/resolve/main/generation_config.json,https://huggingface.co/mistralai/Mistral-7B-v0.1/resolve/main/pytorch_model-00001-of-00002.bin,https://huggingface.co/mistralai/Mistral-7B-v0.1/resolve/main/pytorch_model-00002-of-00002.bin,https://huggingface.co/mistralai/Mistral-7B-v0.1/resolve/main/pytorch_model.bin.index.json,https://huggingface.co/mistralai/Mistral-7B-v0.1/resolve/main/special_tokens_map.json,https://huggingface.co/mistralai/Mistral-7B-v0.1/resolve/main/tokenizer.json,https://huggingface.co/mistralai/Mistral-7B-v0.1/resolve/main/tokenizer.model,https://huggingface.co/mistralai/Mistral-7B-v0.1/resolve/main/tokenizer_config.json",
     "mistralai/Mistral-7B-Instruct-v0.1": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/resolve/main/config.json,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/resolve/main/generation_config.json,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/resolve/main/pytorch_model-00001-of-00002.bin,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/resolve/main/pytorch_model-00002-of-00002.bin,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/resolve/main/pytorch_model.bin.index.json,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/resolve/main/special_tokens_map.json,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/resolve/main/tokenizer.json,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/resolve/main/tokenizer.model,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/resolve/main/tokenizer_config.json",
     "mistralai/Mistral-7B-Instruct-v0.2": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2/resolve/main/config.json,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2/resolve/main/generation_config.json,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2/resolve/main/pytorch_model-00001-of-00003.bin,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2/resolve/main/pytorch_model-00002-of-00003.bin,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2/resolve/main/pytorch_model-00003-of-00003.bin,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2/resolve/main/pytorch_model.bin.index.json,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2/resolve/main/special_tokens_map.json,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2/resolve/main/tokenizer.json,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2/resolve/main/tokenizer.model,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2/resolve/main/tokenizer_config.json",
-
     # huggingface-cli prefixed Models will download using the huggingface-cli tool
     # TODO: Convert all of the MODEL_REPOS with a NamedTuple that includes the install_method
     "huggingface-cli/meta-llama/Meta-Llama-3-8B": "",
@@ -73,7 +72,10 @@ def model_should_run_on_event(model: str, event: str, backend: str) -> bool:
     elif event == "periodic":
         # test llama3 on gpu only, see description in https://github.com/pytorch/torchchat/pull/399 for reasoning
         if backend == "gpu":
-            return model in ["openlm-research/open_llama_7b", "huggingface-cli/meta-llama/Meta-Llama-3-8B"]
+            return model in [
+                "openlm-research/open_llama_7b",
+                "huggingface-cli/meta-llama/Meta-Llama-3-8B",
+            ]
         else:
             return model in ["openlm-research/open_llama_7b"]
     else:

GPTQ.py

@@ -360,7 +360,7 @@ def faster_quant(self, H, W):
         groupsize = self.groupsize
         orig_dtype = W.dtype
         W = W.detach().float()
-        rows, columns = W.shape[0], W.shape[1]
+        _, columns = W.shape[0], W.shape[1]
         device = W.device
 
         if groupsize == -1:

build/builder.py

@@ -150,9 +150,7 @@ def __post_init__(self):
         try:
             from tokenizer.tiktoken import Tokenizer as TiktokenTokenizer
 
-            self.t = TiktokenTokenizer(
-                model_path=str(self.tokenizer_path)
-            )
+            self.t = TiktokenTokenizer(model_path=str(self.tokenizer_path))
             self.is_tiktoken = True
             self.is_sentencepiece = False
             return
@@ -162,9 +160,7 @@ def __post_init__(self):
         try:
             from sentencepiece import SentencePieceProcessor
 
-            self.t = SentencePieceProcessor(
-                model_file=str(self.tokenizer_path)
-            )
+            self.t = SentencePieceProcessor(model_file=str(self.tokenizer_path))
             self.is_tiktoken = False
             self.is_sentencepiece = True
             return

build/convert_torchtune_checkpoint.py

@@ -4,12 +4,12 @@
 # This source code is licensed under the license found in the
 # LICENSE file in the root directory of this source tree.
 
+import logging
 import os
 import re
 import sys
-import logging
 from pathlib import Path
-from typing import Dict, List, Optional
+from typing import Dict, List
 
 import torch
 
@@ -46,7 +46,7 @@ def from_hf(
     merged_result: Dict[str, torch.Tensor],
     num_heads: int = 32,
     num_kv_heads: int = 32,
-    dim: int = 4096
+    dim: int = 4096,
 ) -> Dict[str, torch.Tensor]:
     """
     Utility function which converts the given state_dict from the HF format
@@ -56,7 +56,7 @@ def from_hf(
     """
 
     def permute(w, n_heads):
-        head_dim = dim  // n_heads
+        head_dim = dim // n_heads
         return (
             w.view(n_heads, 2, head_dim // 2, dim)
             .transpose(1, 2)
@@ -114,7 +114,7 @@ def convert_torchtune_checkpoint(
             raise RuntimeError(f"{checkpoint_dir / file} is not a file")
 
     # If the model is already in meta format, simply rename it
-    if checkpoint_format == 'meta':
+    if checkpoint_format == "meta":
         if len(checkpoint_files) > 1:
             raise RuntimeError("Multiple meta format checkpoint files not supported")
 
@@ -127,23 +127,27 @@ def convert_torchtune_checkpoint(
         os.rename(checkpoint_path, Path.joinpath(checkpoint_dir, "model.pth"))
 
     # If the model is in HF format, merge all of the checkpoints and then convert
-    elif checkpoint_format == 'hf':
+    elif checkpoint_format == "hf":
         merged_result = {}
         for file in checkpoint_files:
             state_dict = torch.load(
-                Path.joinpath(checkpoint_dir, file), map_location="cpu", mmap=True, weights_only=True
+                Path.joinpath(checkpoint_dir, file),
+                map_location="cpu",
+                mmap=True,
+                weights_only=True,
             )
             merged_result.update(state_dict)
 
         model_config = MODEL_CONFIGS[model_name]
         final_result = from_hf(merged_result, **model_config)
 
-        print(f"Saving checkpoint to {checkpoint_dir / 'model.pth'}. This may take a while.")
+        print(
+            f"Saving checkpoint to {checkpoint_dir / 'model.pth'}. This may take a while."
+        )
         torch.save(final_result, Path.joinpath(checkpoint_dir, "model.pth"))
         print("Done.")
 
 
-
 if __name__ == "__main__":
     import argparse
 
@@ -155,19 +159,19 @@ def convert_torchtune_checkpoint(
     )
     parser.add_argument(
         "--checkpoint-files",
-        nargs='+',
+        nargs="+",
         required=True,
     )
     parser.add_argument(
         "--checkpoint-format",
         type=str,
         required=True,
-        choices=['meta', 'hf'],
+        choices=["meta", "hf"],
     )
     parser.add_argument(
         "--model-name",
         type=str,
-        choices=['llama2_7B', 'llama3_8B'],
+        choices=["llama2_7B", "llama3_8B"],
     )
 
     args = parser.parse_args()

build/gguf_loader.py

@@ -7,20 +7,17 @@
 
 import copy
 import logging
-import sys
-from pathlib import Path
 from typing import Any
 
 import gguf
 
 import torch
 
-
 from gguf import GGUFValueType
-from .model import ModelArgs, Transformer
 from quantize import pack_scales_and_zeros, WeightOnlyInt4Linear
 
 from build.gguf_util import Q4_0, to_float
+from .model import ModelArgs, Transformer
 
 logger: logging.Logger = logging.getLogger(__name__)
 

build/model.py

@@ -15,7 +15,7 @@
 from torch import Tensor
 from torch.nn import functional as F
 
-from build.utils import find_multiple, get_precision, use_aoti_backend
+from build.utils import find_multiple, get_precision
 
 config_path = Path(f"{str(Path(__file__).parent)}/known_model_params")
 

cli.py

@@ -57,14 +57,17 @@ def add_arguments_for_export(parser):
     # Only export specific options should be here
     _add_arguments_common(parser)
 
+
 def add_arguments_for_list(parser):
     # Only list specific options should be here
     _add_arguments_common(parser)
 
+
 def add_arguments_for_remove(parser):
     # Only remove specific options should be here
     _add_arguments_common(parser)
 
+
 def _add_arguments_common(parser):
     # Model specification. TODO Simplify this.
     # A model can be specified using a positional model name or HuggingFace

config/model_config.py

@@ -77,6 +77,7 @@ def load_model_configs() -> Dict[str, ModelConfig]:
 
     return model_configs
 
+
 def resolve_model_config(model: str) -> ModelConfig:
     model = model.lower()
     # Lazy load model config from JSON.

download.py

@@ -7,13 +7,13 @@
 import shutil
 import urllib.request
 from pathlib import Path
-from typing import Optional, Sequence
+from typing import Optional
 
 from build.convert_hf_checkpoint import convert_hf_checkpoint
 from config.model_config import (
+    load_model_configs,
     ModelConfig,
     ModelDistributionChannel,
-    load_model_configs,
     resolve_model_config,
 )
 
@@ -129,16 +129,14 @@ def list_main(args) -> None:
 
         name_col.append(name)
         aliases_col.append(", ".join(config.aliases))
-        installed_col.append('Yes' if is_downloaded else "")
+        installed_col.append("Yes" if is_downloaded else "")
 
-    cols = {
-        "Model": name_col,
-        "Aliases": aliases_col,
-        "Downloaded": installed_col
-    }
+    cols = {"Model": name_col, "Aliases": aliases_col, "Downloaded": installed_col}
 
     # Find the length of the longest value in each column.
-    col_widths = {key:max(*[len(s) for s in vals], len(key)) + 1 for (key,vals) in cols.items()}
+    col_widths = {
+        key: max(*[len(s) for s in vals], len(key)) + 1 for (key, vals) in cols.items()
+    }
 
     # Display header.
     print()

export.py

@@ -18,7 +18,7 @@
     TokenizerArgs,
 )
 
-from build.utils import set_backend, set_precision, use_aoti_backend, use_et_backend
+from build.utils import set_backend, set_precision
 from cli import add_arguments, add_arguments_for_export, arg_init, check_args
 from export_aoti import export_model as export_model_aoti
 

export_et.py

@@ -11,7 +11,6 @@
 
 from executorch.backends.xnnpack.partition.xnnpack_partitioner import (
     XnnpackDynamicallyQuantizedPartitioner,
-    XnnpackPartitioner,
 )
 
 # TODO: change back to executorch.examples.portable.utils

hqq/core/bitpack.py

@@ -1,9 +1,10 @@
 # Written by Dr. Hicham Badri @Mobius Labs GmbH - 2023
 #####################################################
 
-import torch
-from torch import uint8, int32, Tensor
 import numpy as np
+import torch
+from torch import int32, Tensor, uint8
+
 from .utils import is_divisible
 
 

hqq/core/optimize.py

@@ -1,8 +1,8 @@
 # Written by Dr. Hicham Badri @Mobius Labs GmbH - 2023
 #####################################################
+import numpy as np
 import torch
 from torch import Tensor
-import numpy as np
 
 
 # re-estimate the scale based on the inverse median

hqq/core/peft.py

@@ -1,10 +1,9 @@
 # Written by Dr. Hicham Badri @Mobius Labs GmbH - 2023
 #####################################################
-import torch
-from torch import float16, bfloat16, float32
-from torch import Tensor
-from torch import nn
 import numpy as np
+import torch
+from torch import bfloat16, float16, float32, nn, Tensor
+
 from .quantize import HQQLinear, Quantizer
 from .utils import cleanup
 

hqq/core/quantize.py

@@ -1,13 +1,15 @@
 # Written by Dr. Hicham Badri @Mobius Labs GmbH - 2023
 #####################################################
-import torch
-from torch import uint8, int32, float16, nn, Tensor
 import copy
 from enum import Enum
 
-from .utils import is_divisible
-from .optimize import optimize_weights_proximal
+import torch
+from torch import float16, int32, nn, Tensor, uint8
+
 from .bitpack import BitPack
+from .optimize import optimize_weights_proximal
+
+from .utils import is_divisible
 
 
 # Main HQQ Quantizer

hqq/core/utils.py

@@ -1,8 +1,9 @@
 # Written by Dr. Hicham Badri @Mobius Labs GmbH - 2023
 #####################################################
-import torch
 import gc
+
 import numpy as np
+import torch
 
 
 def cleanup() -> None:

torchchat.py

@@ -34,7 +34,7 @@
         add_help=True,
     )
     # Default command is to print help
-    parser.set_defaults(func=lambda args: self._parser.print_help())
+    parser.set_defaults(func=parser.print_help())
 
     add_arguments(parser)
     subparsers = parser.add_subparsers(