no need to use primary_weights_in_fp8 since already using `fp8_auto…

…cast` around model init to initialize the fp8_meta tensors

Signed-off-by: Sudhakar Singh <[email protected]>

transformer_engine/pytorch/module/base.py

@@ -528,7 +528,7 @@ def fp8_init(self, num_gemms: int = 1) -> None:
         self.fp8_calibration = is_fp8_calibration()
         self.fp8_meta["fp8_checkpoint"] = self.fp8 or self.fp8_calibration
 
-        if self.fp8 or self.fp8_calibration or self.primary_weights_in_fp8:
+        if self.fp8 or self.fp8_calibration:
             # FP8 init has already been run and recipe is the same, don't do anything.
             if self.fp8_initialized and get_fp8_recipe() == self.fp8_meta["recipe"]:
                 return

transformer_engine/pytorch/module/linear.py

@@ -552,8 +552,6 @@ def __init__(
         self.ub_split_rs = ub_split_rs
         self.ub_split_ag = ub_split_ag
 
-        self.primary_weights_in_fp8 = primary_weights_in_fp8
-
         if ub_split_rs or ub_split_ag:
             assert (
                 tex.userbuf_comm_available()