diff --git a/README.md b/README.md
index 504f4798..5d0fe0b3 100644
--- a/README.md
+++ b/README.md
@@ -19,6 +19,12 @@ Results on CNN/Dailymail (25/3/2019):
 
 **Package Requirements**: pytorch pytorch_pretrained_bert tensorboardX multiprocess pyrouge
 
+We have included a requirements.txt file to help you get started. 
+
+```
+pip install -r requirements.txt
+```
+
 Some codes are borrowed from ONMT(https://github.com/OpenNMT/OpenNMT-py)
 
 ## Data Preparation For CNN/Dailymail
@@ -97,4 +103,21 @@ python train.py -mode validate -bert_data_path ../bert_data/cnndm -model_path MO
 * `MODEL_PATH` is the directory of saved checkpoints
 * `RESULT_PATH` is where you want to put decoded summaries (default `../results/cnndm`)
 
+## Known Issues
+
+When running the code with higher versions of PyTorch, you may encounter the following error.
+
+```
+Subtraction, the - operator, with a bool tensor is not supported
+```
+
+For this reason we have specified version 1.1.0 in the requirements.txt file.
+
+When running one of the training commands, the program may crash with the following message.
+
+```
+invalid device ordinal
+```
+
+This may be because you are running on a system that only has access to a single GPU. If this is the case, running the commands with the following flags ``-visible_gpus 0  -gpu_ranks 0 -world_size 1`` should work.
 
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 00000000..45c007c8
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1,23 @@
+boto3==1.16.45
+botocore==1.19.45
+certifi==2020.12.5
+chardet==4.0.0
+dataclasses==0.8
+dill==0.3.3
+idna==2.10
+jmespath==0.10.0
+multiprocess==0.70.11.1
+numpy==1.19.4
+protobuf==3.14.0
+pyrouge==0.1.3
+python-dateutil==2.8.1
+pytorch-pretrained-bert==0.6.2
+regex==2020.11.13
+requests==2.25.1
+s3transfer==0.3.3
+six==1.15.0
+tensorboardX==2.1
+torch==1.1.0
+tqdm==4.55.0
+typing-extensions==3.7.4.3
+urllib3==1.26.2
\ No newline at end of file