diff --git a/fairseq/dataclass/configs.py b/fairseq/dataclass/configs.py
index af957fec64..c6284e75fd 100644
--- a/fairseq/dataclass/configs.py
+++ b/fairseq/dataclass/configs.py
@@ -1126,16 +1126,16 @@ class EMAConfig(FairseqDataclass):
 
 @dataclass
 class FairseqConfig(FairseqDataclass):
-    common: CommonConfig = CommonConfig()
-    common_eval: CommonEvalConfig = CommonEvalConfig()
-    distributed_training: DistributedTrainingConfig = DistributedTrainingConfig()
-    dataset: DatasetConfig = DatasetConfig()
-    optimization: OptimizationConfig = OptimizationConfig()
-    checkpoint: CheckpointConfig = CheckpointConfig()
-    bmuf: FairseqBMUFConfig = FairseqBMUFConfig()
-    generation: GenerationConfig = GenerationConfig()
-    eval_lm: EvalLMConfig = EvalLMConfig()
-    interactive: InteractiveConfig = InteractiveConfig()
+    common: CommonConfig = field(default=CommonConfig)
+    common_eval: CommonEvalConfig = field(default=CommonEvalConfig)
+    distributed_training: DistributedTrainingConfig = field(default=DistributedTrainingConfig)
+    dataset: DatasetConfig = field(default=DatasetConfig)
+    optimization: OptimizationConfig = field(default=OptimizationConfig)
+    checkpoint: CheckpointConfig = field(default=CheckpointConfig)
+    bmuf: FairseqBMUFConfig = field(default=FairseqBMUFConfig)
+    generation: GenerationConfig = field(default=GenerationConfig)
+    eval_lm: EvalLMConfig = field(default=EvalLMConfig)
+    interactive: InteractiveConfig = field(default=InteractiveConfig)
     model: Any = MISSING
     task: Any = None
     criterion: Any = None
@@ -1144,4 +1144,4 @@ class FairseqConfig(FairseqDataclass):
     scoring: Any = None
     bpe: Any = None
     tokenizer: Any = None
-    ema: EMAConfig = EMAConfig()
+    ema: EMAConfig = field(default=EMAConfig)
diff --git a/fairseq/models/transformer/transformer_config.py b/fairseq/models/transformer/transformer_config.py
index 4650de2e17..a6a31a6c5d 100644
--- a/fairseq/models/transformer/transformer_config.py
+++ b/fairseq/models/transformer/transformer_config.py
@@ -111,13 +111,13 @@ class TransformerConfig(FairseqDataclass):
         },
     )
     adaptive_input: bool = False
-    encoder: EncDecBaseConfig = EncDecBaseConfig()
+    encoder: EncDecBaseConfig = field(default=EncDecBaseConfig)
     # TODO should really be in the encoder config
     max_source_positions: int = field(
         default=DEFAULT_MAX_SOURCE_POSITIONS,
         metadata={"help": "Maximum input length supported by the encoder"},
     )
-    decoder: DecoderConfig = DecoderConfig()
+    decoder: DecoderConfig = field(default=DecoderConfig)
     # TODO should really be in the decoder config
     max_target_positions: int = field(
         default=DEFAULT_MAX_TARGET_POSITIONS,
@@ -197,7 +197,7 @@ class TransformerConfig(FairseqDataclass):
         default=False, metadata={"help": "perform cross+self-attention"}
     )
     # args for Training with Quantization Noise for Extreme Model Compression ({Fan*, Stock*} et al., 2020)
-    quant_noise: QuantNoiseConfig = field(default=QuantNoiseConfig())
+    quant_noise: QuantNoiseConfig = field(default=QuantNoiseConfig)
     min_params_to_wrap: int = field(
         default=DEFAULT_MIN_PARAMS_TO_WRAP,
         metadata={
diff --git a/fairseq/version.txt b/fairseq/version.txt
index 26acbf080b..6d1ce3d115 100644
--- a/fairseq/version.txt
+++ b/fairseq/version.txt
@@ -1 +1 @@
-0.12.2
+0.12.3.1
\ No newline at end of file
diff --git a/setup.py b/setup.py
index dae06080c5..8bd07b833e 100644
--- a/setup.py
+++ b/setup.py
@@ -179,8 +179,8 @@ def do_setup(package_data):
         install_requires=[
             "cffi",
             "cython",
-            "hydra-core>=1.0.7,<1.1",
-            "omegaconf<2.1",
+            "hydra-core==1.3.2",
+            "omegaconf==2.3.0",
             "numpy>=1.21.3",
             "regex",
             "sacrebleu>=1.4.12",