ruff fix

06_gpu_and_ml/protein-folding/boltz1.py

@@ -54,9 +54,10 @@ def main(
 
     output_path = here / "data" / "boltz1_result.tar.gz"
     print(f"locally writing output tar to {output_path}")
-    with open(output_path, 'wb') as f:
+    with open(output_path, "wb") as f:
         f.write(output)
 
+
 # ## Installing Boltz-1 Python dependencies on Modal
 
 # Code running on Modal runs inside containers built from [container images](https://modal.com/docs/guide/images)
@@ -69,7 +70,9 @@ def main(
 
 image = (
     modal.Image.debian_slim(python_version="3.12")
-    .run_commands("uv pip install --system --compile-bytecode boltz==0.3.2 hf_transfer==0.1.8")
+    .run_commands(
+        "uv pip install --system --compile-bytecode boltz==0.3.2 hf_transfer==0.1.8"
+    )
     .env({"HF_HUB_ENABLE_HF_TRANSFER": "1"})
 )
 
@@ -89,7 +92,9 @@ def main(
 # A Modal Volume is a file system that all of your code running on Modal (or elsewhere!) can access.
 # For more on storing model weights on Modal, see [this guide](https://modal.com/docs/guide/model-weights).
 
-boltz_model_volume = modal.Volume.from_name("boltz1-models", create_if_missing=True)
+boltz_model_volume = modal.Volume.from_name(
+    "boltz1-models", create_if_missing=True
+)
 models_dir = Path("/models/boltz1")
 
 # ## Running Boltz-1 on Modal
@@ -102,6 +107,7 @@ def main(
 # an amino acid sequence as well as a path to an MSA file. We create these
 # files paths in the function below so they can be picked up by the inference code.
 
+
 @app.function(
     image=image,
     volumes={models_dir: boltz_model_volume},
@@ -112,17 +118,17 @@ def boltz1_inference(msadata, ligandyaml):
     ligand_filename = "ligand.yaml"
     temp_filename = "temp.tar.gz"
 
-    Path('./seq1.a3m').write_text(msadata)
-    Path(f'./{ligand_filename}').write_text(ligandyaml)
+    Path("./seq1.a3m").write_text(msadata)
+    Path(f"./{ligand_filename}").write_text(ligandyaml)
 
     print(f"predicting using boltz model {models_dir}")
-    os.system(f'boltz predict {ligand_filename} --cache={models_dir}')
+    os.system(f"boltz predict {ligand_filename} --cache={models_dir}")
 
     print(f"creating tar file {temp_filename} of outputs")
     os.system(f"tar czvf {temp_filename} boltz_results_ligand")
 
     print(f"converting {temp_filename} to bytes for returning")
-    output = open(f'./{temp_filename}', 'rb').read()
+    output = open(f"./{temp_filename}", "rb").read()
     return output
 
 
@@ -142,6 +148,7 @@ def boltz1_inference(msadata, ligandyaml):
 # as a separate task, concurrently. We don't need to worry about this use of `async`
 # spreading to the rest of our code -- Modal launches just this Function in an async runtime.
 
+
 @app.function(
     volumes={models_dir: boltz_model_volume},
     timeout=20 * MINUTES,
@@ -152,7 +159,7 @@ async def download_inference_dependencies(force=False):
 
     import aiohttp
 
-    base_url = "https://huggingface.co/boltz-community/boltz-1/resolve/e01950840c2a2ec881695f26e994a73b417af0b2/" # sic
+    base_url = "https://huggingface.co/boltz-community/boltz-1/resolve/e01950840c2a2ec881695f26e994a73b417af0b2/"  # sic
     inference_dependencies = [
         "boltz1.ckpt",
         "ccd.pkl",