diff --git a/.github/workflows/doctests.yml b/.github/workflows/doctests.yml
new file mode 100644
index 0000000000..913fb7f4c5
--- /dev/null
+++ b/.github/workflows/doctests.yml
@@ -0,0 +1,42 @@
+name: Doctests
+
+on:
+  push:
+    branches:
+      - doctest*
+  repository_dispatch:
+  schedule:
+    - cron: "0 0 * * *"
+
+
+env:
+  HF_HOME: /mnt/cache
+  TRANSFORMERS_IS_CI: yes
+  RUN_SLOW: yes
+  OMP_NUM_THREADS: 16
+  MKL_NUM_THREADS: 16
+  PYTEST_TIMEOUT: 600
+
+jobs:
+  run_doctests:
+    runs-on: [self-hosted, docker-gpu, single-gpu]
+    container:
+      image: pytorch/pytorch:1.9.0-cuda11.1-cudnn8-runtime
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y libsndfile1-dev
+          pip install --upgrade pip
+          pip install .[dev]
+
+      - name: Run doctests
+        run: |
+          pytest --doctest-modules $(cat utils/documentation_tests.txt) -sv --doctest-continue-on-failure
diff --git a/docs/source/quicktour.rst b/docs/source/quicktour.rst
index 4db1645c09..b0039ff646 100644
--- a/docs/source/quicktour.rst
+++ b/docs/source/quicktour.rst
@@ -65,7 +65,7 @@ make them readable. For instance:
 .. code-block::
 
     >>> classifier('We are very happy to show you the 🤗 Transformers library.')
-    [{'label': 'POSITIVE', 'score': 0.99978}]
+    [{'label': 'POSITIVE', 'score': 0.9998}]
 
 That's encouraging! You can use it on a list of sentences, which will be preprocessed then fed to the model as a
 `batch`, returning a list of dictionaries like this one:
diff --git a/docs/source/task_summary.rst b/docs/source/task_summary.rst
index 0aa0d83412..fc6766ed1b 100644
--- a/docs/source/task_summary.rst
+++ b/docs/source/task_summary.rst
@@ -345,27 +345,27 @@ This outputs the sequences with the mask filled, the confidence score, and the t
 
     >>> from pprint import pprint
     >>> pprint(unmasker(f"HuggingFace is creating a {unmasker.tokenizer.mask_token} that the community uses to solve NLP tasks."))
-    [{'score': 0.179275,
+    [{'score': 0.1793,
       'sequence': 'HuggingFace is creating a tool that the community uses to solve '
                   'NLP tasks.',
       'token': 3944,
       'token_str': ' tool'},
-     {'score': 0.113494,
+     {'score': 0.1135,
       'sequence': 'HuggingFace is creating a framework that the community uses to '
                   'solve NLP tasks.',
       'token': 7208,
       'token_str': ' framework'},
-     {'score': 0.0524355,
+     {'score': 0.0524,
       'sequence': 'HuggingFace is creating a library that the community uses to '
                   'solve NLP tasks.',
       'token': 5560,
       'token_str': ' library'},
-     {'score': 0.0349353,
+     {'score': 0.0349,
       'sequence': 'HuggingFace is creating a database that the community uses to '
                   'solve NLP tasks.',
       'token': 8503,
       'token_str': ' database'},
-     {'score': 0.0286025,
+     {'score': 0.0286,
       'sequence': 'HuggingFace is creating a prototype that the community uses to '
                   'solve NLP tasks.',
       'token': 17715,
@@ -458,7 +458,7 @@ of tokens.
 .. code-block::
 
     >>> ## PYTORCH CODE
-    >>> from transformers import AutoModelForCausalLM, AutoTokenizer, set_seed, top_k_top_p_filtering
+    >>> from transformers import AutoModelForCausalLM, AutoTokenizer, top_k_top_p_filtering
     >>> import torch
     >>> from torch import nn
 
@@ -476,9 +476,6 @@ of tokens.
     >>> # filter
     >>> filtered_next_token_logits = top_k_top_p_filtering(next_token_logits, top_k=50, top_p=1.0)
 
-    >>> # set seed for reproducibility
-    >>> set_seed(42)
-
     >>> # sample
     >>> probs = nn.functional.softmax(filtered_next_token_logits, dim=-1)
     >>> next_token = torch.multinomial(probs, num_samples=1)
@@ -486,8 +483,10 @@ of tokens.
     >>> generated = torch.cat([input_ids, next_token], dim=-1)
 
     >>> resulting_string = tokenizer.decode(generated.tolist()[0])
+    >>> print(resulting_string)
+    Hugging Face is based in DUMBO, New York City, and ...
     >>> ## TENSORFLOW CODE
-    >>> from transformers import TFAutoModelForCausalLM, AutoTokenizer, set_seed, tf_top_k_top_p_filtering
+    >>> from transformers import TFAutoModelForCausalLM, AutoTokenizer, tf_top_k_top_p_filtering
     >>> import tensorflow as tf
 
     >>> tokenizer = AutoTokenizer.from_pretrained("gpt2")
@@ -504,24 +503,17 @@ of tokens.
     >>> # filter
     >>> filtered_next_token_logits = tf_top_k_top_p_filtering(next_token_logits, top_k=50, top_p=1.0)
 
-    >>> # set seed for reproducibility
-    >>> set_seed(42)
-
     >>> # sample
     >>> next_token = tf.random.categorical(filtered_next_token_logits, dtype=tf.int32, num_samples=1)
 
     >>> generated = tf.concat([input_ids, next_token], axis=1)
 
     >>> resulting_string = tokenizer.decode(generated.numpy().tolist()[0])
-
-
-This outputs a (hopefully) coherent next token following the original sequence, which in our case is the word
-*features*:
-
-.. code-block::
-
     >>> print(resulting_string)
-    Hugging Face is based in DUMBO, New York City, and features
+    Hugging Face is based in DUMBO, New York City, and ...
+
+This outputs a (hopefully) coherent next token following the original sequence, which in our case is the word *is* or
+*features*.
 
 In the next section, we show how :func:`~transformers.generation_utils.GenerationMixin.generate` can be used to
 generate multiple tokens up to a specified length instead of one token at a time.
@@ -555,7 +547,7 @@ Below is an example of text generation using ``XLNet`` and its tokenizer, which
 .. code-block::
 
     >>> ## PYTORCH CODE
-    >>> from transformers import AutoModelForCausalLM, AutoTokenizer, set_seed
+    >>> from transformers import AutoModelForCausalLM, AutoTokenizer
 
     >>> model = AutoModelForCausalLM.from_pretrained("xlnet-base-cased")
     >>> tokenizer = AutoTokenizer.from_pretrained("xlnet-base-cased")
@@ -573,19 +565,14 @@ Below is an example of text generation using ``XLNet`` and its tokenizer, which
     ... with people, even a bishop, begging for his blessing. <eod> </s> <eos>"""
 
     >>> prompt = "Today the weather is really nice and I am planning on "
-    >>> inputs = tokenizer.encode(PADDING_TEXT + prompt, add_special_tokens=False, return_tensors="pt")
+    >>> inputs = tokenizer(PADDING_TEXT + prompt, add_special_tokens=False, return_tensors="pt")["input_ids"]
 
-    >>> prompt_length = len(tokenizer.decode(inputs[0], skip_special_tokens=True, clean_up_tokenization_spaces=True))
+    >>> prompt_length = len(tokenizer.decode(inputs[0]))
     >>> outputs = model.generate(inputs, max_length=250, do_sample=True, top_p=0.95, top_k=60)
-    >>> # set seed for reproducibility
-    >>> set_seed(42)
-    >>> generated = prompt + tokenizer.decode(outputs[0])[prompt_length:]
+    >>> generated = prompt + tokenizer.decode(outputs[0])[prompt_length+1:]
 
     >>> print(generated)
-    Today the weather is really nice and I am planning on anning on going to a nearby restaurant on Monday. It is very
-    cool with the clouds and the wind. A nice afternoon is on the way out of there, when I get my phone in the sun.
-    Sounds like its a good day in my house, but on that "good"" thing. There is a group of people who'd want to be out
-    and
+    Today the weather is really nice and I am planning ...
     >>> ## TENSORFLOW CODE
     >>> from transformers import TFAutoModelForCausalLM, AutoTokenizer
 
@@ -605,18 +592,14 @@ Below is an example of text generation using ``XLNet`` and its tokenizer, which
     ... with people, even a bishop, begging for his blessing. <eod> </s> <eos>"""
 
     >>> prompt = "Today the weather is really nice and I am planning on "
-    >>> inputs = tokenizer.encode(PADDING_TEXT + prompt, add_special_tokens=False, return_tensors="tf")
+    >>> inputs = tokenizer(PADDING_TEXT + prompt, add_special_tokens=False, return_tensors="tf")["input_ids"]
 
-    >>> prompt_length = len(tokenizer.decode(inputs[0], skip_special_tokens=True, clean_up_tokenization_spaces=True))
-    >>> # set seed for reproducibility
-    >>> set_seed(42)
+    >>> prompt_length = len(tokenizer.decode(inputs[0]))
     >>> outputs = model.generate(inputs, max_length=250, do_sample=True, top_p=0.95, top_k=60)
-    >>> generated = prompt + tokenizer.decode(outputs[0])[prompt_length:]
+    >>> generated = prompt + tokenizer.decode(outputs[0])[prompt_length+1:]
 
     >>> print(generated)
-    Today the weather is really nice and I am planning on anning on riding a "" over to the coast. It is also an ideal
-    day to fly to Bali and see more of the local "".<eop> “...The weather is great for travel and traveling as far as
-    local "".”. When the weather is good, I will ride my "" over to the coast and see more of
+    Today the weather is really nice and I am planning ...
 
 
 Text generation is currently possible with *GPT-2*, *OpenAi-GPT*, *CTRL*, *XLNet*, *Transfo-XL* and *Reformer* in
@@ -669,18 +652,18 @@ Here are the expected results:
 
     >>> for entity in ner_pipe(sequence):
     ...     print(entity)
-    {'entity': 'I-ORG', 'score': 0.999579, 'index': 1, 'word': 'Hu', 'start': 0, 'end': 2}
-    {'entity': 'I-ORG', 'score': 0.990976, 'index': 2, 'word': '##gging', 'start': 2, 'end': 7}
-    {'entity': 'I-ORG', 'score': 0.998223, 'index': 3, 'word': 'Face', 'start': 8, 'end': 12}
-    {'entity': 'I-ORG', 'score': 0.999488, 'index': 4, 'word': 'Inc', 'start': 13, 'end': 16}
-    {'entity': 'I-LOC', 'score': 0.999434, 'index': 11, 'word': 'New', 'start': 40, 'end': 43}
-    {'entity': 'I-LOC', 'score': 0.999320, 'index': 12, 'word': 'York', 'start': 44, 'end': 48}
-    {'entity': 'I-LOC', 'score': 0.999379, 'index': 13, 'word': 'City', 'start': 49, 'end': 53}
-    {'entity': 'I-LOC', 'score': 0.986258, 'index': 19, 'word': 'D', 'start': 79, 'end': 80}
-    {'entity': 'I-LOC', 'score': 0.951427, 'index': 20, 'word': '##UM', 'start': 80, 'end': 82}
-    {'entity': 'I-LOC', 'score': 0.933659, 'index': 21, 'word': '##BO', 'start': 82, 'end': 84}
-    {'entity': 'I-LOC', 'score': 0.976165, 'index': 28, 'word': 'Manhattan', 'start': 114, 'end': 123}
-    {'entity': 'I-LOC', 'score': 0.991463, 'index': 29, 'word': 'Bridge', 'start': 124, 'end': 130}
+    {'entity': 'I-ORG', 'score': 0.9996, 'index': 1, 'word': 'Hu', 'start': 0, 'end': 2}
+    {'entity': 'I-ORG', 'score': 0.9910, 'index': 2, 'word': '##gging', 'start': 2, 'end': 7}
+    {'entity': 'I-ORG', 'score': 0.9982, 'index': 3, 'word': 'Face', 'start': 8, 'end': 12}
+    {'entity': 'I-ORG', 'score': 0.9995, 'index': 4, 'word': 'Inc', 'start': 13, 'end': 16}
+    {'entity': 'I-LOC', 'score': 0.9994, 'index': 11, 'word': 'New', 'start': 40, 'end': 43}
+    {'entity': 'I-LOC', 'score': 0.9993, 'index': 12, 'word': 'York', 'start': 44, 'end': 48}
+    {'entity': 'I-LOC', 'score': 0.9994, 'index': 13, 'word': 'City', 'start': 49, 'end': 53}
+    {'entity': 'I-LOC', 'score': 0.9863, 'index': 19, 'word': 'D', 'start': 79, 'end': 80}
+    {'entity': 'I-LOC', 'score': 0.9514, 'index': 20, 'word': '##UM', 'start': 80, 'end': 82}
+    {'entity': 'I-LOC', 'score': 0.9337, 'index': 21, 'word': '##BO', 'start': 82, 'end': 84}
+    {'entity': 'I-LOC', 'score': 0.9762, 'index': 28, 'word': 'Manhattan', 'start': 114, 'end': 123}
+    {'entity': 'I-LOC', 'score': 0.9915, 'index': 29, 'word': 'Bridge', 'start': 124, 'end': 130}
 
 Note how the tokens of the sequence "Hugging Face" have been identified as an organisation, and "New York City",
 "DUMBO" and "Manhattan Bridge" have been identified as locations.
diff --git a/utils/documentation_tests.txt b/utils/documentation_tests.txt
new file mode 100644
index 0000000000..5aaed9f481
--- /dev/null
+++ b/utils/documentation_tests.txt
@@ -0,0 +1,2 @@
+docs/source/quicktour.rst
+docs/source/task_summary.rst
\ No newline at end of file