[devops] remove post commit ci (#5566)

* [devops] remove post commit ci

* [misc] run pre-commit on all files

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

examples/images/vit/vit_benchmark.py

@@ -119,9 +119,7 @@ def main():
             if hasattr(booster.plugin, "stage_manager") and booster.plugin.stage_manager is not None:
                 # run pipeline forward backward
                 batch = iter([batch])
-                outputs = booster.execute_pipeline(
-                    batch, model, criterion, optimizer, return_loss=True
-                )
+                outputs = booster.execute_pipeline(batch, model, criterion, optimizer, return_loss=True)
             else:
                 outputs = model(**batch)
                 loss = criterion(outputs, None)

examples/language/data_utils.py

@@ -121,4 +121,4 @@ class RandomDataset(Dataset):
             "input_ids": self.input_ids[idx],
             "attention_mask": self.attention_mask[idx],
             "labels": self.input_ids[idx],
-        }
+        }

examples/language/llama2/finetune.py

@@ -270,9 +270,7 @@ def main():
         ) as pbar:
             for step in pbar:
                 if use_pipeline:
-                    outputs = booster.execute_pipeline(
-                        dataloader_iter, model, _criterion, optimizer, return_loss=True
-                    )
+                    outputs = booster.execute_pipeline(dataloader_iter, model, _criterion, optimizer, return_loss=True)
                     loss = outputs["loss"]
                 else:
                     batch = next(dataloader_iter)

examples/language/llama2/pretrain.py

@@ -285,9 +285,7 @@ def main():
         ) as pbar:
             for step in pbar:
                 if use_pipeline:
-                    outputs = booster.execute_pipeline(
-                        dataloader_iter, model, _criterion, optimizer, return_loss=True
-                    )
+                    outputs = booster.execute_pipeline(dataloader_iter, model, _criterion, optimizer, return_loss=True)
                     loss = outputs["loss"]
                 else:
                     batch = next(dataloader_iter)

examples/language/openmoe/benchmark/utils.py

@@ -50,7 +50,6 @@ def all_reduce_mean(x: float, world_size: int) -> float:
 
 
 class Timer:
-
     def __init__(self) -> None:
         self.start_time: Optional[float] = None
         self.duration: float = 0.0
@@ -112,7 +111,7 @@ class PerformanceEvaluator:
         batch_size, seq_len = input_ids.shape
 
         self.num_samples += batch_size
-        self.flop += (batch_size * seq_len * self.model_numel * 2 * (3 + int(self.enable_grad_checkpoint)))
+        self.flop += batch_size * seq_len * self.model_numel * 2 * (3 + int(self.enable_grad_checkpoint))
 
     def on_fit_end(self) -> None:
         avg_duration = all_reduce_mean(self.timer.duration, self.world_size)
@@ -122,5 +121,6 @@ class PerformanceEvaluator:
         if dist.get_rank() == 0:
             print(
                 f"num_samples: {self.num_samples}, dp_world_size: {self.dp_world_size}, flop: {self.flop}, avg_duration: {avg_duration}, "
-                f"avg_throughput: {avg_throughput}")
+                f"avg_throughput: {avg_throughput}"
+            )
             print(f"Throughput: {avg_throughput:.2f} samples/sec, TFLOPS per GPU: {avg_tflops_per_gpu:.2f}")

examples/language/openmoe/infer.py

@@ -16,17 +16,15 @@ def inference(args):
     tokenizer = T5Tokenizer.from_pretrained("google/umt5-small")
     if args.model == "test":
         config = LlamaConfig.from_pretrained("hpcai-tech/openmoe-base")
-        set_openmoe_args(config,
-                         num_experts=config.num_experts,
-                         moe_layer_interval=config.moe_layer_interval,
-                         enable_kernel=True)
+        set_openmoe_args(
+            config, num_experts=config.num_experts, moe_layer_interval=config.moe_layer_interval, enable_kernel=True
+        )
         model = OpenMoeForCausalLM(config)
     else:
         config = LlamaConfig.from_pretrained(f"hpcai-tech/openmoe-{args.model}")
-        set_openmoe_args(config,
-                         num_experts=config.num_experts,
-                         moe_layer_interval=config.moe_layer_interval,
-                         enable_kernel=False)
+        set_openmoe_args(
+            config, num_experts=config.num_experts, moe_layer_interval=config.moe_layer_interval, enable_kernel=False
+        )
         model = OpenMoeForCausalLM.from_pretrained(f"hpcai-tech/openmoe-{args.model}", config=config)
     model = model.eval().bfloat16()
     model = model.to(torch.cuda.current_device())

examples/language/openmoe/model/convert_openmoe_ckpt.py

@@ -172,9 +172,9 @@ def make_state_dict(converted_params):
 def load_t5x_weights_in_t5(model, config, t5x_checkpoint_path):
     """Replaces the params in model witht the T5X converted params."""
     variables = checkpoints.load_t5x_checkpoint(t5x_checkpoint_path)
-    converted = convert_t5x_to_pytorch(variables,
-                                       num_layers=config.num_hidden_layers,
-                                       moe_interval=config.moe_layer_interval)
+    converted = convert_t5x_to_pytorch(
+        variables, num_layers=config.num_hidden_layers, moe_interval=config.moe_layer_interval
+    )
     state_dict = make_state_dict(converted)
     model.load_state_dict(state_dict, strict=True)
 
@@ -203,11 +203,9 @@ def convert_t5x_checkpoint_to_pytorch(t5x_checkpoint_path, config_file, pytorch_
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="Converts a native T5X checkpoint into a PyTorch checkpoint.")
     # Required parameters
-    parser.add_argument("--t5x_checkpoint_path",
-                        default=None,
-                        type=str,
-                        required=True,
-                        help="Path to the T5X checkpoint.")
+    parser.add_argument(
+        "--t5x_checkpoint_path", default=None, type=str, required=True, help="Path to the T5X checkpoint."
+    )
     parser.add_argument(
         "--config_file",
         default=None,
@@ -215,10 +213,8 @@ if __name__ == "__main__":
         required=True,
         help="The config json file corresponding to the pre-trained T5 model.\nThis specifies the model architecture.",
     )
-    parser.add_argument("--pytorch_dump_path",
-                        default=None,
-                        type=str,
-                        required=True,
-                        help="Path to the output PyTorch model.")
+    parser.add_argument(
+        "--pytorch_dump_path", default=None, type=str, required=True, help="Path to the output PyTorch model."
+    )
     args = parser.parse_args()
     convert_t5x_checkpoint_to_pytorch(args.t5x_checkpoint_path, args.config_file, args.pytorch_dump_path)

examples/language/opt/opt_train_demo.py

@@ -41,9 +41,7 @@ def train_epoch(epoch, model, optimizer, _criterion, lr_scheduler, dataloader, b
         # Forward pass
         for _ in pbar:
             if use_pipeline:
-                outputs = booster.execute_pipeline(
-                    dataloader, model, _criterion, optimizer, return_loss=True
-                )
+                outputs = booster.execute_pipeline(dataloader, model, _criterion, optimizer, return_loss=True)
                 # Backward and optimize
                 if is_pp_last_stage:
                     loss = outputs["loss"]