diff --git a/src/conv_lstm_layer.c b/src/conv_lstm_layer.c
index b51a3aed..12426ab6 100644
--- a/src/conv_lstm_layer.c
+++ b/src/conv_lstm_layer.c
@@ -64,65 +64,65 @@ layer make_conv_lstm_layer(int batch, int h, int w, int c, int output_filters, i
     l.peephole = peephole;
 
     // U
-    l.uf = (layer*)malloc(sizeof(layer));
+    l.uf = (layer*)calloc(1, sizeof(layer));
     *(l.uf) = make_convolutional_layer(batch, steps, h, w, c, output_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
     l.uf->batch = batch;
     if (l.workspace_size < l.uf->workspace_size) l.workspace_size = l.uf->workspace_size;
 
-    l.ui = (layer*)malloc(sizeof(layer));
+    l.ui = (layer*)calloc(1, sizeof(layer));
     *(l.ui) = make_convolutional_layer(batch, steps, h, w, c, output_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
     l.ui->batch = batch;
     if (l.workspace_size < l.ui->workspace_size) l.workspace_size = l.ui->workspace_size;
 
-    l.ug = (layer*)malloc(sizeof(layer));
+    l.ug = (layer*)calloc(1, sizeof(layer));
     *(l.ug) = make_convolutional_layer(batch, steps, h, w, c, output_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
     l.ug->batch = batch;
     if (l.workspace_size < l.ug->workspace_size) l.workspace_size = l.ug->workspace_size;
 
-    l.uo = (layer*)malloc(sizeof(layer));
+    l.uo = (layer*)calloc(1, sizeof(layer));
     *(l.uo) = make_convolutional_layer(batch, steps, h, w, c, output_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
     l.uo->batch = batch;
     if (l.workspace_size < l.uo->workspace_size) l.workspace_size = l.uo->workspace_size;
 
 
     // W
-    l.wf = (layer*)malloc(sizeof(layer));
+    l.wf = (layer*)calloc(1, sizeof(layer));
     *(l.wf) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
     l.wf->batch = batch;
     if (l.workspace_size < l.wf->workspace_size) l.workspace_size = l.wf->workspace_size;
 
-    l.wi = (layer*)malloc(sizeof(layer));
+    l.wi = (layer*)calloc(1, sizeof(layer));
     *(l.wi) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
     l.wi->batch = batch;
     if (l.workspace_size < l.wi->workspace_size) l.workspace_size = l.wi->workspace_size;
 
-    l.wg = (layer*)malloc(sizeof(layer));
+    l.wg = (layer*)calloc(1, sizeof(layer));
     *(l.wg) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
     l.wg->batch = batch;
     if (l.workspace_size < l.wg->workspace_size) l.workspace_size = l.wg->workspace_size;
 
-    l.wo = (layer*)malloc(sizeof(layer));
+    l.wo = (layer*)calloc(1, sizeof(layer));
     *(l.wo) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
     l.wo->batch = batch;
     if (l.workspace_size < l.wo->workspace_size) l.workspace_size = l.wo->workspace_size;
 
 
     // V
-    l.vf = (layer*)malloc(sizeof(layer));
+    l.vf = (layer*)calloc(1, sizeof(layer));
     if (l.peephole) {
         *(l.vf) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
         l.vf->batch = batch;
         if (l.workspace_size < l.vf->workspace_size) l.workspace_size = l.vf->workspace_size;
     }
 
-    l.vi = (layer*)malloc(sizeof(layer));
+    l.vi = (layer*)calloc(1, sizeof(layer));
     if (l.peephole) {
         *(l.vi) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
         l.vi->batch = batch;
         if (l.workspace_size < l.vi->workspace_size) l.workspace_size = l.vi->workspace_size;
     }
 
-    l.vo = (layer*)malloc(sizeof(layer));
+    l.vo = (layer*)calloc(1, sizeof(layer));
     if (l.peephole) {
         *(l.vo) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
         l.vo->batch = batch;
@@ -909,7 +909,7 @@ void forward_conv_lstm_layer_gpu(layer l, network_state state)
         fix_nan_and_inf(l.h_gpu, l.outputs*l.batch);
 
         copy_ongpu(l.outputs*l.batch, l.c_gpu, 1, l.cell_gpu, 1);
-        copy_ongpu(l.outputs*l.batch, l.h_gpu, 1, l.output_gpu, 1); // required for both Detection and Training
+        copy_ongpu(l.outputs*l.batch, l.h_gpu, 1, l.output_gpu, 1); // is required for both Detection and Training
 
         state.input += l.inputs*l.batch;
         l.output_gpu    += l.outputs*l.batch;
diff --git a/src/crnn_layer.c b/src/crnn_layer.c
index f466508e..8534c69a 100644
--- a/src/crnn_layer.c
+++ b/src/crnn_layer.c
@@ -48,17 +48,17 @@ layer make_crnn_layer(int batch, int h, int w, int c, int hidden_filters, int ou
 
     l.state = (float*)calloc(l.hidden * l.batch * (l.steps + 1), sizeof(float));
 
-    l.input_layer = (layer*)malloc(sizeof(layer));
+    l.input_layer = (layer*)calloc(1, sizeof(layer));
     *(l.input_layer) = make_convolutional_layer(batch, steps, h, w, c, hidden_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
     l.input_layer->batch = batch;
     if (l.workspace_size < l.input_layer->workspace_size) l.workspace_size = l.input_layer->workspace_size;
 
-    l.self_layer = (layer*)malloc(sizeof(layer));
+    l.self_layer = (layer*)calloc(1, sizeof(layer));
     *(l.self_layer) = make_convolutional_layer(batch, steps, h, w, hidden_filters, hidden_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
     l.self_layer->batch = batch;
     if (l.workspace_size < l.self_layer->workspace_size) l.workspace_size = l.self_layer->workspace_size;
 
-    l.output_layer = (layer*)malloc(sizeof(layer));
+    l.output_layer = (layer*)calloc(1, sizeof(layer));
     *(l.output_layer) = make_convolutional_layer(batch, steps, h, w, hidden_filters, output_filters, groups, size, stride, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
     l.output_layer->batch = batch;
     if (l.workspace_size < l.output_layer->workspace_size) l.workspace_size = l.output_layer->workspace_size;
diff --git a/src/lstm_layer.c b/src/lstm_layer.c
index f7b79477..94664ce3 100644
--- a/src/lstm_layer.c
+++ b/src/lstm_layer.c
@@ -39,49 +39,49 @@ layer make_lstm_layer(int batch, int inputs, int outputs, int steps, int batch_n
     l.out_h = 1;
     l.out_c = outputs;
 
-    l.uf = (layer*)malloc(sizeof(layer));
+    l.uf = (layer*)calloc(1, sizeof(layer));
     fprintf(stderr, "\t\t");
     *(l.uf) = make_connected_layer(batch, steps, inputs, outputs, LINEAR, batch_normalize);
     l.uf->batch = batch;
     if (l.workspace_size < l.uf->workspace_size) l.workspace_size = l.uf->workspace_size;
 
-    l.ui = (layer*)malloc(sizeof(layer));
+    l.ui = (layer*)calloc(1, sizeof(layer));
     fprintf(stderr, "\t\t");
     *(l.ui) = make_connected_layer(batch, steps, inputs, outputs, LINEAR, batch_normalize);
     l.ui->batch = batch;
     if (l.workspace_size < l.ui->workspace_size) l.workspace_size = l.ui->workspace_size;
 
-    l.ug = (layer*)malloc(sizeof(layer));
+    l.ug = (layer*)calloc(1, sizeof(layer));
     fprintf(stderr, "\t\t");
     *(l.ug) = make_connected_layer(batch, steps, inputs, outputs, LINEAR, batch_normalize);
     l.ug->batch = batch;
     if (l.workspace_size < l.ug->workspace_size) l.workspace_size = l.ug->workspace_size;
 
-    l.uo = (layer*)malloc(sizeof(layer));
+    l.uo = (layer*)calloc(1, sizeof(layer));
     fprintf(stderr, "\t\t");
     *(l.uo) = make_connected_layer(batch, steps, inputs, outputs, LINEAR, batch_normalize);
     l.uo->batch = batch;
     if (l.workspace_size < l.uo->workspace_size) l.workspace_size = l.uo->workspace_size;
 
-    l.wf = (layer*)malloc(sizeof(layer));
+    l.wf = (layer*)calloc(1, sizeof(layer));
     fprintf(stderr, "\t\t");
     *(l.wf) = make_connected_layer(batch, steps, outputs, outputs, LINEAR, batch_normalize);
     l.wf->batch = batch;
     if (l.workspace_size < l.wf->workspace_size) l.workspace_size = l.wf->workspace_size;
 
-    l.wi = (layer*)malloc(sizeof(layer));
+    l.wi = (layer*)calloc(1, sizeof(layer));
     fprintf(stderr, "\t\t");
     *(l.wi) = make_connected_layer(batch, steps, outputs, outputs, LINEAR, batch_normalize);
     l.wi->batch = batch;
     if (l.workspace_size < l.wi->workspace_size) l.workspace_size = l.wi->workspace_size;
 
-    l.wg = (layer*)malloc(sizeof(layer));
+    l.wg = (layer*)calloc(1, sizeof(layer));
     fprintf(stderr, "\t\t");
     *(l.wg) = make_connected_layer(batch, steps, outputs, outputs, LINEAR, batch_normalize);
     l.wg->batch = batch;
     if (l.workspace_size < l.wg->workspace_size) l.workspace_size = l.wg->workspace_size;
 
-    l.wo = (layer*)malloc(sizeof(layer));
+    l.wo = (layer*)calloc(1, sizeof(layer));
     fprintf(stderr, "\t\t");
     *(l.wo) = make_connected_layer(batch, steps, outputs, outputs, LINEAR, batch_normalize);
     l.wo->batch = batch;
diff --git a/src/parser.c b/src/parser.c
index 95c92250..63a22319 100644
--- a/src/parser.c
+++ b/src/parser.c
@@ -258,7 +258,7 @@ layer parse_conv_lstm(list *options, size_params params)
     ACTIVATION activation = get_activation(activation_s);
     int batch_normalize = option_find_int_quiet(options, "batch_normalize", 0);
     int xnor = option_find_int_quiet(options, "xnor", 0);
-    int peephole = option_find_int_quiet(options, "peephole", 1);
+    int peephole = option_find_int_quiet(options, "peephole", 0);
 
     layer l = make_conv_lstm_layer(params.batch, params.h, params.w, params.c, output_filters, groups, params.time_steps, size, stride, padding, activation, batch_normalize, peephole, xnor);
 
diff --git a/src/rnn_layer.c b/src/rnn_layer.c
index 28163d75..4b5b9c2c 100644
--- a/src/rnn_layer.c
+++ b/src/rnn_layer.c
@@ -42,19 +42,19 @@ layer make_rnn_layer(int batch, int inputs, int hidden, int outputs, int steps,
 
     l.state = (float*)calloc(batch * hidden * (steps + 1), sizeof(float));
 
-    l.input_layer = (layer*)malloc(sizeof(layer));
+    l.input_layer = (layer*)calloc(1, sizeof(layer));
     fprintf(stderr, "\t\t");
     *(l.input_layer) = make_connected_layer(batch, steps, inputs, hidden, activation, batch_normalize);
     l.input_layer->batch = batch;
     if (l.workspace_size < l.input_layer->workspace_size) l.workspace_size = l.input_layer->workspace_size;
 
-    l.self_layer = (layer*)malloc(sizeof(layer));
+    l.self_layer = (layer*)calloc(1, sizeof(layer));
     fprintf(stderr, "\t\t");
     *(l.self_layer) = make_connected_layer(batch, steps, hidden, hidden, (log==2)?LOGGY:(log==1?LOGISTIC:activation), batch_normalize);
     l.self_layer->batch = batch;
     if (l.workspace_size < l.self_layer->workspace_size) l.workspace_size = l.self_layer->workspace_size;
 
-    l.output_layer = (layer*)malloc(sizeof(layer));
+    l.output_layer = (layer*)calloc(1, sizeof(layer));
     fprintf(stderr, "\t\t");
     *(l.output_layer) = make_connected_layer(batch, steps, hidden, outputs, activation, batch_normalize);
     l.output_layer->batch = batch;