diff --git a/bin/run-ldc93s1.sh b/bin/run-ldc93s1.sh
index b71d5a24..4bee5c70 100755
--- a/bin/run-ldc93s1.sh
+++ b/bin/run-ldc93s1.sh
@@ -16,6 +16,10 @@ else
     checkpoint_dir=$(python -c 'from xdg import BaseDirectory as xdg; print(xdg.save_data_path("deepspeech/ldc93s1"))')
 fi
 
+# Force only one visible device because we have a single-sample dataset
+# and when trying to run on multiple devices (like GPUs), this will break
+export CUDA_VISIBLE_DEVICES=0
+
 python -u DeepSpeech.py --noshow_progressbar \
   --train_files data/ldc93s1/ldc93s1.csv \
   --test_files data/ldc93s1/ldc93s1.csv \
diff --git a/bin/run-tc-ldc93s1_checkpoint.sh b/bin/run-tc-ldc93s1_checkpoint.sh
index e37e92ae..565d6986 100755
--- a/bin/run-tc-ldc93s1_checkpoint.sh
+++ b/bin/run-tc-ldc93s1_checkpoint.sh
@@ -12,6 +12,10 @@ if [ ! -f "${ldc93s1_dir}/ldc93s1.csv" ]; then
     python -u bin/import_ldc93s1.py ${ldc93s1_dir}
 fi;
 
+# Force only one visible device because we have a single-sample dataset
+# and when trying to run on multiple devices (like GPUs), this will break
+export CUDA_VISIBLE_DEVICES=0
+
 python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
   --train_files ${ldc93s1_csv} --train_batch_size 1 \
   --dev_files ${ldc93s1_csv} --dev_batch_size 1 \
diff --git a/bin/run-tc-ldc93s1_new.sh b/bin/run-tc-ldc93s1_new.sh
index 73fc2558..64fbed76 100755
--- a/bin/run-tc-ldc93s1_new.sh
+++ b/bin/run-tc-ldc93s1_new.sh
@@ -12,6 +12,10 @@ if [ ! -f "${ldc93s1_dir}/ldc93s1.csv" ]; then
     python -u bin/import_ldc93s1.py ${ldc93s1_dir}
 fi;
 
+# Force only one visible device because we have a single-sample dataset
+# and when trying to run on multiple devices (like GPUs), this will break
+export CUDA_VISIBLE_DEVICES=0
+
 python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
   --train_files ${ldc93s1_csv} --train_batch_size 1 \
   --feature_cache '/tmp/ldc93s1_cache' \
diff --git a/bin/run-tc-ldc93s1_singleshotinference.sh b/bin/run-tc-ldc93s1_singleshotinference.sh
index 4ded226a..0db2a202 100755
--- a/bin/run-tc-ldc93s1_singleshotinference.sh
+++ b/bin/run-tc-ldc93s1_singleshotinference.sh
@@ -10,6 +10,10 @@ if [ ! -f "${ldc93s1_dir}/ldc93s1.csv" ]; then
     python -u bin/import_ldc93s1.py ${ldc93s1_dir}
 fi;
 
+# Force only one visible device because we have a single-sample dataset
+# and when trying to run on multiple devices (like GPUs), this will break
+export CUDA_VISIBLE_DEVICES=0
+
 python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
   --train_files ${ldc93s1_csv} --train_batch_size 1 \
   --dev_files ${ldc93s1_csv} --dev_batch_size 1 \
diff --git a/bin/run-tc-ldc93s1_tflite.sh b/bin/run-tc-ldc93s1_tflite.sh
index e2e8fb61..bab6d7b0 100755
--- a/bin/run-tc-ldc93s1_tflite.sh
+++ b/bin/run-tc-ldc93s1_tflite.sh
@@ -10,6 +10,10 @@ if [ ! -f "${ldc93s1_dir}/ldc93s1.csv" ]; then
     python -u bin/import_ldc93s1.py ${ldc93s1_dir}
 fi;
 
+# Force only one visible device because we have a single-sample dataset
+# and when trying to run on multiple devices (like GPUs), this will break
+export CUDA_VISIBLE_DEVICES=0
+
 python -u DeepSpeech.py --noshow_progressbar \
   --n_hidden 100 \
   --checkpoint_dir '/tmp/ckpt' \