remove fused_layernorm with apex dependency and include low_precision…

…_layernorm
mosaicml · Jan 8, 2024 · c44effa · c44effa
1 parent cb78d90
commit c44effa
Show file tree

Hide file tree

Showing 5 changed files with 5 additions and 5 deletions.
diff --git a/examples/benchmarks/bert/yamls/finetuning/glue/mcloud_run.yaml b/examples/benchmarks/bert/yamls/finetuning/glue/mcloud_run.yaml
@@ -83,7 +83,7 @@ parameters:
 
   # Algorithms
   algorithms:
-    fused_layernorm: {}
+    low_precision_layernorm: {}
 
   # Task configuration
   tasks:

diff --git a/examples/benchmarks/bert/yamls/finetuning/glue/mosaic-bert-base-uncased.yaml b/examples/benchmarks/bert/yamls/finetuning/glue/mosaic-bert-base-uncased.yaml
@@ -47,7 +47,7 @@ scheduler:
 
 # Algorithms
 algorithms:
-  fused_layernorm: {}
+  low_precision_layernorm: {}
 
 # Task configuration
 tasks:

diff --git a/examples/benchmarks/bert/yamls/main/mcloud_run_a100_40gb.yaml b/examples/benchmarks/bert/yamls/main/mcloud_run_a100_40gb.yaml
@@ -108,7 +108,7 @@ parameters:
     weight_decay: 1.0e-5 # Amount of weight decay regularization
 
   algorithms:
-    fused_layernorm: {}
+    low_precision_layernorm: {}
 
   max_duration: 286720000sp # Subsample the training data for ~275M samples
   eval_interval: 2000ba

diff --git a/examples/benchmarks/bert/yamls/main/mcloud_run_a100_80gb.yaml b/examples/benchmarks/bert/yamls/main/mcloud_run_a100_80gb.yaml
@@ -108,7 +108,7 @@ parameters:
     weight_decay: 1.0e-5 # Amount of weight decay regularization
 
   algorithms:
-    fused_layernorm: {}
+    low_precision_layernorm: {}
 
   max_duration: 286720000sp # Subsample the training data for ~275M samples
   eval_interval: 2000ba

diff --git a/examples/benchmarks/bert/yamls/main/mosaic-bert-base-uncased.yaml b/examples/benchmarks/bert/yamls/main/mosaic-bert-base-uncased.yaml
@@ -69,7 +69,7 @@ optimizer:
   weight_decay: 1.0e-5 # Amount of weight decay regularization
 
 algorithms:
-  fused_layernorm: {}
+  low_precision_layernorm: {}
 
 max_duration: 286720000sp # Subsample the training data for ~275M samples
 eval_interval: 2000ba