Add CLI launch options for new args, add new args to LaunchOptions, rename to , use shorthand convention for --bind in singularity command

Author: XkunW

vec_inf/cli/_cli.py

@@ -72,6 +72,21 @@ def cli() -> None:
     type=str,
     help="Quality of service",
 )
+@click.option(
+    "--exclude",
+    type=str,
+    help="Exclude certain nodes from the resources granted to the job",
+)
+@click.option(
+    "--node-list",
+    type=str,
+    help="Request a specific list of nodes for deployment",
+)
+@click.option(
+    "--bind",
+    type=str,
+    help="Additional binds for the singularity container as a comma separated list of bind paths",
+)
 @click.option(
     "--time",
     type=str,
@@ -124,8 +139,16 @@ def launch(
             Number of nodes to use
         - gpus_per_node : int, optional
             Number of GPUs per node
+        - account : str, optional
+            Charge resources used by this job to specified account
         - qos : str, optional
             Quality of service tier
+        - exclude : str, optional
+            Exclude certain nodes from the resources granted to the job
+        - node_list : str, optional
+            Request a specific list of nodes for deployment
+        - bind : str, optional
+            Additional binds for the singularity container
         - time : str, optional
             Time limit for job
         - venv : str, optional

vec_inf/client/_client_vars.py

@@ -166,7 +166,7 @@ class SlurmScriptTemplate(TypedDict):
         f"export LD_LIBRARY_PATH={LD_LIBRARY_PATH}",
         f"export VLLM_NCCL_SO_PATH={VLLM_NCCL_SO_PATH}",
     ],
-    "singularity_command": f"singularity exec --nv --bind {{model_weights_path}}:{{model_weights_path}}{{additional_binds}} --containall {SINGULARITY_IMAGE}",
+    "singularity_command": f"singularity exec --nv --bind {{model_weights_path}}{{additional_binds}} --containall {SINGULARITY_IMAGE}",
     "activate_venv": "source {venv}/bin/activate",
     "server_setup": {
         "single_node": [

vec_inf/client/_slurm_script_generator.py

@@ -39,7 +39,7 @@ def __init__(self, params: dict[str, Any]):
         self.params = params
         self.is_multinode = int(self.params["num_nodes"]) > 1
         self.use_singularity = self.params["venv"] == "singularity"
-        self.additional_binds = self.params.get("binds", "")
+        self.additional_binds = self.params.get("bind", "")
         if self.additional_binds:
             self.additional_binds = f" --bind {self.additional_binds}"
         self.model_weights_path = str(

vec_inf/client/config.py

@@ -114,7 +114,7 @@ class ModelConfig(BaseModel):
     node_list: Optional[str] = Field(
         default=None, description="Request a specific list of nodes for deployment"
     )
-    binds: Optional[str] = Field(
+    bind: Optional[str] = Field(
         default=None, description="Additional binds for the singularity container"
     )
     venv: str = Field(

vec_inf/client/models.py

@@ -170,6 +170,12 @@ class LaunchOptions:
         Quality of Service level
     time : str, optional
         Time limit for the job
+    exclude : str, optional
+        Exclude certain nodes from the resources granted to the job
+    node_list : str, optional
+        Request a specific list of nodes for deployment
+    bind : str, optional
+        Additional binds for the singularity container
     vocab_size : int, optional
         Size of model vocabulary
     data_type : str, optional
@@ -191,6 +197,9 @@ class LaunchOptions:
     gpus_per_node: Optional[int] = None
     account: Optional[str] = None
     qos: Optional[str] = None
+    exclude: Optional[str] = None
+    node_list: Optional[str] = None
+    bind: Optional[str] = None
     time: Optional[str] = None
     vocab_size: Optional[int] = None
     data_type: Optional[str] = None