diff --git a/deepspeed/__init__.py b/deepspeed/__init__.py
index de5d1efcd1aa..a6c906f1119a 100755
--- a/deepspeed/__init__.py
+++ b/deepspeed/__init__.py
@@ -4,16 +4,18 @@
 import sys
 import types
 
-from deepspeed.runtime.engine import DeepSpeedEngine
-from deepspeed.runtime.engine import ADAM_OPTIMIZER, LAMB_OPTIMIZER
-from deepspeed.runtime.lr_schedules import add_tuning_arguments
-from deepspeed.runtime.config import DeepSpeedConfig
-from deepspeed.runtime.activation_checkpointing import checkpointing
-from deepspeed.ops.transformer import DeepSpeedTransformerLayer, DeepSpeedTransformerConfig
-from deepspeed.utils import logger
+from . import ops
+
+from .runtime.engine import DeepSpeedEngine
+from .runtime.engine import ADAM_OPTIMIZER, LAMB_OPTIMIZER
+from .runtime.lr_schedules import add_tuning_arguments
+from .runtime.config import DeepSpeedConfig
+from .runtime.activation_checkpointing import checkpointing
+from .ops.transformer import DeepSpeedTransformerLayer, DeepSpeedTransformerConfig
+from .utils import logger
 
 try:
-    from deepspeed.git_version_info import version, git_hash, git_branch
+    from .git_version_info import version, git_hash, git_branch
 except ImportError:
     version = "0.0.0+unknown"
     git_hash = None
diff --git a/deepspeed/ops/__init__.py b/deepspeed/ops/__init__.py
index e69de29bb2d1..6c4187415aae 100644
--- a/deepspeed/ops/__init__.py
+++ b/deepspeed/ops/__init__.py
@@ -0,0 +1,7 @@
+from ..git_version_info import installed_ops as __installed_ops__
+from . import lamb
+from . import transformer
+if __installed_ops__['sparse-attn']:
+    from . import sparse_attention
+if __installed_ops__['cpu-adam']:
+    from . import adam
diff --git a/deepspeed/ops/adam/__init__.py b/deepspeed/ops/adam/__init__.py
index 13883dc7b590..1d8844409374 100755
--- a/deepspeed/ops/adam/__init__.py
+++ b/deepspeed/ops/adam/__init__.py
@@ -1 +1 @@
-from deepspeed.ops.adam.cpu_adam import DeepSpeedCPUAdam
+from .cpu_adam import DeepSpeedCPUAdam
diff --git a/deepspeed/ops/adam/cpu_adam.py b/deepspeed/ops/adam/cpu_adam.py
index 5387090e3815..888be4dc9ad4 100755
--- a/deepspeed/ops/adam/cpu_adam.py
+++ b/deepspeed/ops/adam/cpu_adam.py
@@ -1,7 +1,8 @@
 import math
 import torch
-import deepspeed.ops.adam.cpu_adam_op as ds_opt_adam
+import importlib
 
+ds_opt_adam = None
 
 class DeepSpeedCPUAdam(torch.optim.Optimizer):
 
@@ -24,6 +25,9 @@ def __init__(self,
         super(DeepSpeedCPUAdam, self).__init__(model_params, default_args)
         self.opt_id = DeepSpeedCPUAdam.optimizer_id
         DeepSpeedCPUAdam.optimizer_id = DeepSpeedCPUAdam.optimizer_id + 1
+
+        global ds_opt_adam
+        ds_opt_adam = importlib.import_module('deepspeed.ops.adam.cpu_adam_op')
         ds_opt_adam.create_adam(self.opt_id, lr, bettas[0], bettas[1], eps, weight_decay)
 
     def __setstate__(self, state):
diff --git a/setup.py b/setup.py
index 02f2f7fd005a..ae5010a65b11 100755
--- a/setup.py
+++ b/setup.py
@@ -27,6 +27,12 @@ def fetch_requirements(path):
 dev_requires = fetch_requirements('requirements/requirements-dev.txt')
 sparse_attn_requires = fetch_requirements('requirements/requirements-sparse-attn.txt')
 
+# Constants for each op
+LAMB = "lamb"
+TRANSFORMER = "transformer"
+SPARSE_ATTN = "sparse-attn"
+ADAM = "cpu-adam"
+
 # Build environment variables for custom builds
 DS_BUILD_LAMB_MASK = 1
 DS_BUILD_TRANSFORMER_MASK = 10
@@ -50,15 +56,15 @@ def fetch_requirements(path):
 BUILD_MASK = (DS_BUILD_LAMB | DS_BUILD_TRANSFORMER | DS_BUILD_SPARSE_ATTN
               | DS_BUILD_ADAM)
 
-install_ops = []
+install_ops = dict.fromkeys([LAMB, TRANSFORMER, SPARSE_ATTN, ADAM], False)
 if BUILD_MASK & DS_BUILD_LAMB:
-    install_ops.append('lamb')
+    install_ops[LAMB] = True
 if BUILD_MASK & DS_BUILD_ADAM:
-    install_ops.append('adam')
+    install_ops[ADAM] = True
 if BUILD_MASK & DS_BUILD_TRANSFORMER:
-    install_ops.append('transformer')
+    install_ops[TRANSFORMER] = True
 if BUILD_MASK & DS_BUILD_SPARSE_ATTN:
-    install_ops.append('sparse-attn')
+    install_ops[SPARSE_ATTN] = True
 if len(install_ops) == 0:
     print("Building without any cuda/cpp extensions")
 print(f'BUILD_MASK={BUILD_MASK}, install_ops={install_ops}')
@@ -196,14 +202,21 @@ def fetch_requirements(path):
 
 
 def command_exists(cmd):
-    result = subprocess.Popen(f'type {cmd}', stdout=subprocess.PIPE, shell=True)
-    return result.wait() == 0
+    if '|' in cmd:
+        cmds = cmd.split("|")
+    else:
+        cmds = [cmd]
+    valid = False
+    for cmd in cmds:
+        result = subprocess.Popen(f'type {cmd}', stdout=subprocess.PIPE, shell=True)
+        valid = valid or result.wait() == 0
+    return valid
 
 
 ## Sparse transformer ##
 if BUILD_MASK & DS_BUILD_SPARSE_ATTN:
     # Check to see if llvm and cmake are installed since they are dependencies
-    required_commands = ['llc-9', 'cmake']
+    required_commands = ['llvm-config|llvm-config-9', 'cmake']
 
     command_status = list(map(command_exists, required_commands))
     if not all(command_status):
@@ -213,6 +226,8 @@ def command_exists(cmd):
         )
         warnings.warn(
             'Skipping sparse attention installation due to missing required packages')
+        # remove from installed ops list
+        install_ops[SPARSE_ATTN] = False
     elif TORCH_MAJOR == 1 and TORCH_MINOR >= 5:
         ext_modules.append(
             CppExtension(name='deepspeed.ops.sparse_attention.cpp_utils',
@@ -223,6 +238,8 @@ def command_exists(cmd):
         install_requires += sparse_attn_requires
     else:
         warnings.warn('Unable to meet requirements to install sparse attention')
+        # remove from installed ops list
+        install_ops[SPARSE_ATTN] = False
 
 # Add development requirements
 install_requires += dev_requires
@@ -243,6 +260,7 @@ def command_exists(cmd):
     fd.write(f"version='{VERSION}+{git_hash}'\n")
     fd.write(f"git_hash='{git_hash}'\n")
     fd.write(f"git_branch='{git_branch}'\n")
+    fd.write(f"installed_ops={install_ops}\n")
 
 print(f'install_requires={install_requires}')
 
diff --git a/tests/unit/adam_test.py b/tests/unit/adam_test.py
index 8e6caa382c12..0fedfeb18e42 100755
--- a/tests/unit/adam_test.py
+++ b/tests/unit/adam_test.py
@@ -1,5 +1,5 @@
 import torch
-from deepspeed import DeepSpeedCPUAdam
+from deepspeed.ops.adam import DeepSpeedCPUAdam
 import time
 
 device = 'cpu'
diff --git a/tests/unit/test_adam_acuracy.py b/tests/unit/test_adam_acuracy.py
index 410543f516d1..b002231622f4 100755
--- a/tests/unit/test_adam_acuracy.py
+++ b/tests/unit/test_adam_acuracy.py
@@ -6,10 +6,10 @@
 import pytest
 import copy
 
-from deepspeed import DeepSpeedCPUAdam
+from deepspeed.ops.adam import DeepSpeedCPUAdam
 
 def check_equal(first, second, atol=1e-2, verbose=False):
-    if verbos:
+    if verbose:
         print(first)
         print(second)
     x = first.detach().numpy()