Fix fp8_buf for Linear and LayerNormLinear (#1633)

* Fix fp8_buf for Linear and LayerNormLinear

Signed-off-by: Kirthi Shankar Sivamani <[email protected]>

* Fix

Signed-off-by: Kirthi Shankar Sivamani <[email protected]>

---------

Signed-off-by: Kirthi Shankar Sivamani <[email protected]>

Author: ksivaman

transformer_engine/pytorch/module/layernorm_linear.py

@@ -1262,6 +1262,7 @@ def forward(
         inp: torch.Tensor,
         is_first_microbatch: Optional[bool] = None,
         fp8_output: Optional[bool] = False,
+        fp8_grad: Optional[bool] = False,
     ) -> Union[torch.Tensor, Tuple[torch.Tensor, ...]]:
         """
         Apply layer normalization to the input followed by a linear transformation.
@@ -1292,6 +1293,13 @@ def forward(
         if skip_fp8_weight_update is not None:
             is_first_microbatch = False
 
+        if self.ub_overlap_rs_fprop:
+            if get_ub(self.ub_name + "_fprop").is_fp8_ubuf():
+                fp8_output = True
+        if self.ub_overlap_rs_dgrad:
+            if get_ub(self.ub_name + "_dgrad").is_fp8_ubuf():
+                fp8_grad = True
+
         with self.prepare_forward(
             inp, allow_non_contiguous=False  # removed .contiguous from inside the layer
         ) as inp:
@@ -1319,7 +1327,7 @@ def forward(
                 output_quantizer,
                 grad_output_quantizer,
                 grad_input_quantizer,
-            ) = self._get_quantizers(fp8_output)
+            ) = self._get_quantizers(fp8_output, fp8_grad)
 
             if torch.is_grad_enabled():
                 fwd_fn = _LayerNormLinear.apply
@@ -1384,7 +1392,7 @@ def forward(
             return out, ln_out
         return out
 
-    def _get_quantizers(self, fp8_output):
+    def _get_quantizers(self, fp8_output, fp8_grad):
         if not self.fp8:
             return [None] * 5
         grad_input_quantizer = None
@@ -1399,6 +1407,8 @@ def _get_quantizers(self, fp8_output):
         if torch.is_grad_enabled():
             grad_output_quantizer = self.quantizers["scaling_bwd"][tex.FP8BwdTensors.GRAD_OUTPUT1]
             grad_output_quantizer.internal = True
+            if fp8_grad:
+                grad_input_quantizer = self.quantizers["scaling_bwd"][tex.FP8BwdTensors.GRAD_INPUT1]
 
         return (
             input_quantizer,

transformer_engine/pytorch/module/layernorm_mlp.py

@@ -1436,6 +1436,11 @@ def forward(
         if skip_fp8_weight_update is not None:
             is_first_microbatch = False
 
+        fp8_output = False
+        if self.ub_overlap_rs:
+            if get_ub("fc2_fprop").is_fp8_ubuf():
+                fp8_output = True
+
         with self.prepare_forward(inp, num_gemms=2) as inp:
             # Get quantizers
             (
@@ -1447,7 +1452,7 @@ def forward(
                 grad_fc1_output_quantizer,
                 grad_fc2_output_quantizer,
                 grad_input_quantizer,
-            ) = self._get_quantizers()
+            ) = self._get_quantizers(fp8_output)
 
             # Get weight tensors
             fc1_weight = self.fc1_weight
@@ -1533,7 +1538,7 @@ def forward(
             return out, ln_out
         return out
 
-    def _get_quantizers(self):
+    def _get_quantizers(self, fp8_output):
         (
             fc1_input_quantizer,
             fc1_weight_quantizer,
@@ -1555,6 +1560,8 @@ def _get_quantizers(self):
             )
             fc2_weight_quantizer = self.quantizers["scaling_fwd"][tex.FP8FwdTensors.GEMM2_WEIGHT]
             fc2_weight_quantizer.internal = True
+            if fp8_output:
+                output_quantizer = self.quantizers["scaling_fwd"][tex.FP8FwdTensors.GEMM2_OUTPUT]
             if torch.is_grad_enabled():
                 grad_fc2_output_quantizer = self.quantizers["scaling_bwd"][
                     tex.FP8BwdTensors.GRAD_OUTPUT1

transformer_engine/pytorch/module/linear.py

@@ -1104,6 +1104,13 @@ def forward(
         if skip_fp8_weight_update is not None:
             is_first_microbatch = False
 
+        if self.ub_overlap_rs_fprop:
+            if get_ub(self.ub_name + "_fprop").is_fp8_ubuf():
+                fp8_output = True
+        if self.ub_overlap_rs_dgrad:
+            if get_ub(self.ub_name + "_dgrad").is_fp8_ubuf():
+                fp8_grad = True
+
         with self.prepare_forward(
             inp,
             allow_non_contiguous=isinstance(inp, QuantizedTensor),