Run black/isort to fix linting (#117)

Summary:

Test Plan:

Author: EntilZha

README.md

@@ -93,7 +93,7 @@ From here there are two options: (1) load weights in our train script and (2) lo
 In your terminal:
 
 ```bash
-python -m bytelatent.hf load-transformers --entropy-repo facebook/blt-entropy --blt-repo facebook/blt-1b hub --prompt "My test prompt"
+python -m bytelatent.hf load-transformers --entropy-repo facebook/blt-entropy --blt-repo facebook/blt-1b --prompt "My test prompt" hub
 ```
 
 In your own code:

bytelatent/config_parser.py

@@ -1,5 +1,5 @@
 import copy
-from typing import Type, TypeVar, Any
+from typing import Any, Type, TypeVar
 
 import omegaconf
 from omegaconf import DictConfig, OmegaConf
@@ -69,15 +69,19 @@ def parse_args_with_default(
 
 T = TypeVar("T", bound=BaseModel)
 
+
 def get_pydantic_default_args(args_cls: Type[T]) -> dict[str, Any]:
     defaults = {}
     for field, info in args_cls.model_fields.items():
         if info.default != PydanticUndefined:
             defaults[field] = info.default
     return defaults
 
+
 def parse_args_to_pydantic_model(
-    args_cls: Type[T], cli_args: DictConfig | None = None, instantiate_default_cls: bool = True
+    args_cls: Type[T],
+    cli_args: DictConfig | None = None,
+    instantiate_default_cls: bool = True,
 ) -> T:
     if instantiate_default_cls:
         default_cfg = OmegaConf.create(args_cls().model_dump())

bytelatent/stool.py

@@ -26,8 +26,10 @@ class StoolArgs(BaseModel):
     dirs_exists_ok: bool = (
         False  # Wether to copy new code and config and run regardless that dir exists
     )
-    override: bool = False  # Whether to delete dump dir and restart, requires confirmation
-    force_override: bool = False # Does not require interaction
+    override: bool = (
+        False  # Whether to delete dump dir and restart, requires confirmation
+    )
+    force_override: bool = False  # Does not require interaction
     nodes: int = -1  # The number of nodes to run the job on.
     ngpu: int = 8  # The number of GPUs required per node.
     ncpu: int = 16  # The number of CPUs allocated per GPU.
@@ -43,7 +45,6 @@ class StoolArgs(BaseModel):
     dry_run: bool = False
 
 
-
 def copy_dir(input_dir: str, output_dir: str) -> None:
     print(f"Copying : {input_dir}\n" f"to      : {output_dir} ...")
     assert os.path.isdir(input_dir), f"{input_dir} is not a directory"
@@ -130,7 +131,9 @@ def launch_job(args: StoolArgs):
     job_name = args.name or args.model_conf["name"]
     dump_dir = os.path.join(args.dump_dir, job_name) or args.model_conf["dump_dir"]
     print("Creating directories...")
-    os.makedirs(dump_dir, exist_ok=args.dirs_exists_ok or args.override or args.force_override)
+    os.makedirs(
+        dump_dir, exist_ok=args.dirs_exists_ok or args.override or args.force_override
+    )
     if args.override or args.force_override:
         if args.force_override:
             shutil.rmtree(dump_dir)
@@ -161,10 +164,10 @@ def launch_job(args: StoolArgs):
         else ""
     )
     env = jinja2.Environment(
-        loader=jinja2.PackageLoader('bytelatent'),
+        loader=jinja2.PackageLoader("bytelatent"),
         autoescape=jinja2.select_autoescape(),
     )
-    template = env.get_template('stool_template.sh.jinja')
+    template = env.get_template("stool_template.sh.jinja")
     sbatch_jinja = template.render(
         name=job_name,
         script=args.script,

demo.py

@@ -11,8 +11,8 @@
 
 
 def main(prompt: str, model_name: str = "blt-1b"):
-    assert model_name in ['blt-1b', 'blt-7b']
-    model_name = model_name.replace('-', '_')
+    assert model_name in ["blt-1b", "blt-7b"]
+    model_name = model_name.replace("-", "_")
     distributed_args = DistributedArgs()
     distributed_args.configure_world()
     if not torch.distributed.is_initialized():
@@ -27,7 +27,9 @@ def main(prompt: str, model_name: str = "blt-1b"):
     patcher_args = train_cfg.data.patcher_args.model_copy(deep=True)
     patcher_args.realtime_patching = True
     print("Loading entropy model and patcher")
-    patcher_args.entropy_model_checkpoint_dir = os.path.join("hf-weights", "entropy_model")
+    patcher_args.entropy_model_checkpoint_dir = os.path.join(
+        "hf-weights", "entropy_model"
+    )
     patcher = patcher_args.build()
     prompts = [prompt]
     outputs = generate_nocache(

pyproject.toml

@@ -46,7 +46,9 @@ pre_build = [
 ]
 compile_xformers = ['xformers']
 dev = [
+    "black==24.8.0",
     "ipython>=9.2.0",
+    "isort>=6.0.1",
     "pudb>=2025.1",
 ]