---
title: "BNNSDirectApplyLSTMBatchBackward(_:_:_:_:_:)"
framework: accelerate
role: symbol
role_heading: Function
path: "accelerate/bnnsdirectapplylstmbatchbackward(_:_:_:_:_:)"
---

# BNNSDirectApplyLSTMBatchBackward(_:_:_:_:_:)

Applies a long short-term memory (LSTM) filter backward to generate gradients.

## Declaration

```swift
func BNNSDirectApplyLSTMBatchBackward(_ layer_params: UnsafePointer<BNNSLayerParametersLSTM>, _ layer_delta_params: UnsafePointer<BNNSLayerParametersLSTM>, _ filter_params: UnsafePointer<BNNSFilterParameters>?, _ training_cache_ptr: UnsafeRawPointer?, _ training_cache_capacity: Int) -> Int32
```

## Parameters

- `layer_params`: The LSTM layer parameters.
- `layer_delta_params`: The LSTM layer delta parameters.
- `filter_params`: Filter runtime parameters.
- `training_cache_ptr`: A pointer to the training cache buffer.
- `training_cache_capacity`: The minimum bytes capacity of the training cache buffer as computed by the training cache capacity function.

## See Also

### Recurrent layers

- [Using Long Short-Term Memory Layers (LSTM)](accelerate/using-long-short-term-memory-layers-lstm.md)
- [BNNSLSTMDataDescriptor](accelerate/bnnslstmdatadescriptor.md)
- [BNNSLSTMGateDescriptor](accelerate/bnnslstmgatedescriptor.md)
- [BNNSLayerFlags](accelerate/bnnslayerflags.md)
- [BNNSLayerParametersLSTM](accelerate/bnnslayerparameterslstm.md)
- [BNNSComputeLSTMTrainingCacheCapacity(_:)](accelerate/bnnscomputelstmtrainingcachecapacity(_:).md)
- [BNNSDirectApplyLSTMBatchTrainingCaching(_:_:_:_:)](accelerate/bnnsdirectapplylstmbatchtrainingcaching(_:_:_:_:).md)