environ.get

docs/cli/profile.md

@@ -4,28 +4,39 @@
 memory footprint and utilization on CPU/GPU.
 Currently, the nos profiler itself runs natively in the execution environment (i.e. outside
 the NOS server), so you'll need to install both the `server` and `test` dependencies alongside
-your existing NOS installation 
+your existing NOS installation.
 
 ```bash
 pip install torch-nos[server,test]
 ```
 
+## NOS Profile Commands
+
+### all
+Profile all models
+
 If you have the time, you can construct a profiling catalog on your machine in its entirety with:
 ```bash
+nos profile all
+```
+### rebuild-catalog
+Generate a fresh catalog with 
+```bash
 nos profile rebuild-catalog
 ```
-
+### model
 You can also profile specific models with 
 ```bash
-nos profile method -m openai/clip-vit-large-patch14
+nos profile model -m openai/clip-vit-large-patch14
 ```
-
+### method
 Or an entire method type with
 ```bash
 nos profile method -m encode_image
 ```
 to benchmark e.g. all image embedding models.
 
+### list
 Dump the nos profiling catalog with `nos profile list`
 
-![NOS Profile List](../assets/nos_profile_list.png)
+![NOS Profile List](../assets/nos_catalog_t4_complete.png)

nos/common/spec.py

@@ -444,7 +444,7 @@ def load_profile_catalog(self) -> "ModelSpecMetadataCatalog":
         catalog_path = NOS_PROFILE_CATALOG_PATH
 
         # Check if we have NOS_PROFILE_CATALOG_PATH_OVERRIDE in the environment
-        if "NOS_PROFILE_CATALOG_PATH_OVERRIDE" in os.environ:
+        if os.environ.get("NOS_PROFILE_CATALOG_PATH_OVERRIDE"):
             catalog_path = Path(os.environ["NOS_PROFILE_CATALOG_PATH_OVERRIDE"])
             logger.debug(f"Using custom profile catalog [path={catalog_path}]")