Merge branch 'zenflow_zero1_2' of github.com:Antlera/DeepSpeed into zenflow_zero1_2

Author: Antlera

deepspeed/compile/profilers/graph_profile.py

@@ -96,6 +96,7 @@ def run(self, *args) -> Any:
         args: inputs to the graph. Tensors in the inpusts must be real tensors, not fake tensors. args can contain ds parameters.
         returns: The output of the graph. Tensor in the output is real tensors.
         """
+        return_val = None
         try:
             assert _all_real_if_tensor(args), "Inputs must be real tensors"
             self.nz3.enable_profiling(True)
@@ -242,6 +243,7 @@ def __init__(self, gm: GraphModule, debug_log=False):
         self.debug_log = debug_log
 
     def run(self, *args) -> Any:
+        return_val = None
         try:
             assert _all_real_if_tensor(args), "Inputs must be real tensors"
             self.nz3.enable_profiling(True)

op_builder/builder.py

@@ -83,8 +83,7 @@ def get_default_compute_capabilities():
 cuda_minor_mismatch_ok = {
     10: ["10.0", "10.1", "10.2"],
     11: ["11.0", "11.1", "11.2", "11.3", "11.4", "11.5", "11.6", "11.7", "11.8"],
-    12: ["12.0", "12.1", "12.2", "12.3", "12.4", "12.5", "12.6",
-         "12.8"],  # There does not appear to be a CUDA Toolkit 12.7
+    12: ["12.0", "12.1", "12.2", "12.3", "12.4", "12.5", "12.6", "12.8", "12.9"],  # There is no CUDATk 12.7
 }