fix(lint): Resolve Pylint errors in train.py and quantizations_test.py

Author: hsuan-lun-chiang

src/maxtext/trainers/pre_train/train.py

@@ -349,7 +349,9 @@ def train_step(model, config, state_mesh_shardings, params_shardings, state, dat
           is_train=True,
       )
     else:
-      OverwriteWithGradient = nnx.variablelib.variable_type_from_name(maxtext_utils.OVERWRITE_WITH_GRADIENT, allow_register=True)
+      OverwriteWithGradient = nnx.variablelib.variable_type_from_name(
+          maxtext_utils.OVERWRITE_WITH_GRADIENT, allow_register=True
+      )
       model_graphdef, curr_params, overwrite_vars, rest = nnx.split(state.model, nnx.Param, OverwriteWithGradient, ...)
       if config.parameter_memory_host_offload:
         # Params are kept on host (pinned_host) in in_shardings. Move only Param
@@ -379,7 +381,9 @@ def diff_wrapper(param, overwrite_vars, rest, config, data):
         return loss, (aux, new_overwrite_vars, new_rest)
 
       grad_func = jax.value_and_grad(diff_wrapper, argnums=(0, 1), has_aux=True)
-      (loss, (aux, new_overwrite_vars, new_rest)), (raw_grads, overwrite_grads) = grad_func(curr_params, overwrite_vars, rest, config, data)
+      (loss, (aux, _, new_rest)), (raw_grads, overwrite_grads) = grad_func(
+          curr_params, overwrite_vars, rest, config, data
+      )
       nnx.update(state.model, new_rest)
       nnx.update(state.model, overwrite_grads)
 

tests/unit/quantizations_test.py

@@ -22,6 +22,7 @@
 from aqt.jax.v2 import aqt_tensor
 from aqt.jax.v2.flax import aqt_flax
 from flax import nnx
+from flax.nnx import traversals
 import jax
 from jax import lax
 from jax import numpy as jnp
@@ -48,7 +49,7 @@ def __init__(
       self,
       quantization: quantizations.AqtQuantization,
       data_type: Any,
-      rngs: nnx.Rngs,
+      rngs: nnx.Rngs,  # pylint: disable=unused-argument
   ):
     self.quantization = quantization
     self.identity = jnp.identity(2, dtype=data_type)
@@ -441,10 +442,9 @@ def loss_quant(model):
 
       # nnx.grad returns a State object which is a mapping of paths to gradients.
       # Flatten them to check for tolerance.
-      from flax.nnx import traversals
       grads_base_flat = traversals.flatten_mapping(grads_base)
       grads_quant_flat = traversals.flatten_mapping(grads_quant)
-      
+
       # Filter for param collections to compare only parameters and not stats/buffers if any
       # Note: NNX grads structure might contain variables like 'kernel', 'bias'.
       # For simplicity we compare all matching keys.
@@ -557,7 +557,7 @@ def test_fp8_gpu_quantization(self):
 
   # @pytest.mark.gpu_only
   @pytest.mark.external_serving
-  def test_fp8_gpu_quantization(self):
+  def test_fp8_gpu_quantization_nnx(self):
     self.quantization_config("fp8_gpu", grad_tolerance=1.5, enable_nnx=True, pure_nnx_decoder=True, pure_nnx=True)
 
   @pytest.mark.gpu_only