diff --git a/TPU_ISSUES_RECORD.md b/TPU_ISSUES_RECORD.md
index 51f5412..74e6503 100644
--- a/TPU_ISSUES_RECORD.md
+++ b/TPU_ISSUES_RECORD.md
@@ -226,6 +226,31 @@ All TPU training issues have been systematically identified and fixed:
 
 **Ready for TPU training test** with 687M parameter brain-to-text model.
 
+---
+
+## New Issue: TPU Memory Exhaustion (2025-10-12 15:00)
+```
+RuntimeError: Bad StatusOr access: RESOURCE_EXHAUSTED: Error allocating device buffer: Attempting to allocate 3.50M. That was not possible. There are 2.07M free.; (0x0x0_HBM0)
+```
+
+**Root Cause**: TPU HBM memory fragmentation with batch_size=64
+- Single batch: 64 × (512 features × 14 patches) × 2 bytes = ~917KB per batch
+- Combined with 687M model parameters + gradients + activations → memory exhaustion
+- TPU memory allocation is stricter than GPU, requires contiguous blocks
+
+**Solution**: Memory-optimized configuration
+```yaml
+# rnn_args.yaml optimizations:
+batch_size: 32  # reduced from 64
+gradient_accumulation_steps: 2  # maintains effective batch size of 64
+num_dataloader_workers: 0  # TPU compatibility
+```
+
+**Memory Calculation**:
+- New batch memory: 32 × 7168 × 2 bytes = ~458KB (50% reduction)
+- Gradient accumulation maintains training stability
+- Effective batch size unchanged: 2 steps × 32 = 64 samples
+
 ## Lessons Learned
 - **Root Cause**: TPU XLA compiler requires strict dtype consistency across all tensors
 - **Key Insight**: `torch.eye()` and `torch.zeros()` default to f32 - must explicitly specify dtype
diff --git a/model_training_nnn/rnn_args.yaml b/model_training_nnn/rnn_args.yaml
index 54c4f79..f0e6847 100644
--- a/model_training_nnn/rnn_args.yaml
+++ b/model_training_nnn/rnn_args.yaml
@@ -21,7 +21,7 @@ use_amp: true # whether to use automatic mixed precision (AMP) for training
 # TPU and distributed training settings
 use_tpu: true # whether to use TPU for training (set to true for TPU)
 num_tpu_cores: 8 # number of TPU cores to use (full TPU v3-8)
-gradient_accumulation_steps: 1 # number of gradient accumulation steps for distributed training
+gradient_accumulation_steps: 2 # number of gradient accumulation steps for distributed training (2x32=64 effective batch size)
 
 output_dir: trained_models/baseline_rnn # directory to save the trained model and logs
 checkpoint_dir: trained_models/baseline_rnn/checkpoint # directory to save checkpoints during training
@@ -75,13 +75,12 @@ dataset:
     smooth_kernel_std: 2 # standard deviation of the smoothing kernel applied to the data
 
   neural_dim: 512 # dimensionality of the neural data
-  batch_size: 64 # batch size for training
+  batch_size: 32 # batch size for training (reduced for TPU memory constraints)
   n_classes: 41 # number of classes (phonemes) in the dataset
   max_seq_elements: 500 # maximum number of sequence elements (phonemes) for any trial
   days_per_batch: 4 # number of randomly-selected days to include in each batch
   seed: 1 # random seed for reproducibility
-  num_dataloader_workers: 4 # number of workers for the data loader
-  dataloader_num_workers: 0 # set to 0 for TPU to avoid multiprocessing issues
+  num_dataloader_workers: 0 # set to 0 for TPU to avoid multiprocessing issues
   loader_shuffle: false # whether to shuffle the data loader
   must_include_days: null # specific days to include in the dataset
   test_percentage: 0.1 # percentage of data to use for testing