diff --git a/examples/README.md b/examples/README.md
index be8b3669f0..afd93716be 100644
--- a/examples/README.md
+++ b/examples/README.md
@@ -135,21 +135,21 @@ Fine-tuning the library models for sequence classification on the GLUE benchmark
Evaluation](https://gluebenchmark.com/). This script can fine-tune the following models: BERT, XLM, XLNet and RoBERTa.
GLUE is made up of a total of 9 different tasks. We get the following results on the dev set of the benchmark with an
-uncased BERT base model (the checkpoint `bert-base-uncased`). All experiments ran on 8 V100 GPUs with a total train
-batch size of 24. Some of these tasks have a small dataset and training can lead to high variance in the results
+uncased BERT base model (the checkpoint `bert-base-uncased`). All experiments ran single V100 GPUs with a total train
+batch sizes between 16 and 64. Some of these tasks have a small dataset and training can lead to high variance in the results
between different runs. We report the median on 5 runs (with different seeds) for each of the metrics.
| Task | Metric | Result |
|-------|------------------------------|-------------|
-| CoLA | Matthew's corr | 48.87 |
-| SST-2 | Accuracy | 91.74 |
-| MRPC | F1/Accuracy | 90.70/86.27 |
-| STS-B | Person/Spearman corr. | 91.39/91.04 |
-| QQP | Accuracy/F1 | 90.79/87.66 |
-| MNLI | Matched acc./Mismatched acc. | 83.70/84.83 |
-| QNLI | Accuracy | 89.31 |
-| RTE | Accuracy | 71.43 |
-| WNLI | Accuracy | 43.66 |
+| CoLA | Matthew's corr | 49.23 |
+| SST-2 | Accuracy | 91.97 |
+| MRPC | F1/Accuracy | 89.47/85.29 |
+| STS-B | Person/Spearman corr. | 83.95/83.70 |
+| QQP | Accuracy/F1 | 88.40/84.31 |
+| MNLI | Matched acc./Mismatched acc. | 80.61/81.08 |
+| QNLI | Accuracy | 87.46 |
+| RTE | Accuracy | 61.73 |
+| WNLI | Accuracy | 45.07 |
Some of these results are significantly different from the ones reported on the test set
of GLUE benchmark on the website. For QQP and WNLI, please refer to [FAQ #12](https://gluebenchmark.com/faq) on the webite.
diff --git a/examples/distillation/README.md b/examples/distillation/README.md
index 24a1677db1..c8fbb01aa4 100644
--- a/examples/distillation/README.md
+++ b/examples/distillation/README.md
@@ -2,23 +2,25 @@
This folder contains the original code used to train Distil* as well as examples showcasing how to use DistilBERT, DistilRoBERTa and DistilGPT2.
-**December 6th, 2019 - Update** We release **DistilmBERT**: 92% of `bert-base-multilingual-cased` on XNLI. The model supports 104 different languages listed [here](https://github.com/google-research/bert/blob/master/multilingual.md#list-of-languages).
+**January 20, 2020 - Bug fixing** We have recently discovered and fixed [a bug](https://github.com/huggingface/transformers/commit/48cbf267c988b56c71a2380f748a3e6092ccaed3) in the evaluation of our `run_*.py` scripts that caused the reported metrics to be over-estimated on average. We have updated all the metrics with the latest runs.
-**November 19th, 2019 - Update** We release German **DistilBERT**: 98.8% of `bert-base-german-dbmdz-cased` on NER tasks.
+**December 6, 2019 - Update** We release **DistilmBERT**: 92% of `bert-base-multilingual-cased` on XNLI. The model supports 104 different languages listed [here](https://github.com/google-research/bert/blob/master/multilingual.md#list-of-languages).
-**October 23rd, 2019 - Update** We release **DistilRoBERTa**: 95% of `RoBERTa-base`'s performance on GLUE, twice as fast as RoBERTa while being 35% smaller.
+**November 19, 2019 - Update** We release German **DistilBERT**: 98.8% of `bert-base-german-dbmdz-cased` on NER tasks.
-**October 3rd, 2019 - Update** We release our [NeurIPS workshop paper](https://arxiv.org/abs/1910.01108) explaining our approach on **DistilBERT**. It includes updated results and further experiments. We applied the same method to GPT2 and release the weights of **DistilGPT2**. DistilGPT2 is two times faster and 33% smaller than GPT2. **The paper superseeds our [previous blogpost](https://medium.com/huggingface/distilbert-8cf3380435b5) with a different distillation loss and better performances. Please use the paper as a reference when comparing/reporting results on DistilBERT.**
+**October 23, 2019 - Update** We release **DistilRoBERTa**: 95% of `RoBERTa-base`'s performance on GLUE, twice as fast as RoBERTa while being 35% smaller.
-**September 19th, 2019 - Update:** We fixed bugs in the code and released an upadted version of the weights trained with a modification of the distillation loss. DistilBERT now reaches 97% of `BERT-base`'s performance on GLUE, and 86.9 F1 score on SQuAD v1.1 dev set (compared to 88.5 for `BERT-base`). We will publish a formal write-up of our approach in the near future!
+**October 3, 2019 - Update** We release our [NeurIPS workshop paper](https://arxiv.org/abs/1910.01108) explaining our approach on **DistilBERT**. It includes updated results and further experiments. We applied the same method to GPT2 and release the weights of **DistilGPT2**. DistilGPT2 is two times faster and 33% smaller than GPT2. **The paper superseeds our [previous blogpost](https://medium.com/huggingface/distilbert-8cf3380435b5) with a different distillation loss and better performances. Please use the paper as a reference when comparing/reporting results on DistilBERT.**
+
+**September 19, 2019 - Update:** We fixed bugs in the code and released an upadted version of the weights trained with a modification of the distillation loss. DistilBERT now reaches 99% of `BERT-base`'s performance on GLUE, and 86.9 F1 score on SQuAD v1.1 dev set (compared to 88.5 for `BERT-base`). We will publish a formal write-up of our approach in the near future!
## What is Distil*
-Distil* is a class of compressed models that started with DistilBERT. DistilBERT stands for Distillated-BERT. DistilBERT is a small, fast, cheap and light Transformer model based on Bert architecture. It has 40% less parameters than `bert-base-uncased`, runs 60% faster while preserving 97% of BERT's performances as measured on the GLUE language understanding benchmark. DistilBERT is trained using knowledge distillation, a technique to compress a large model called the teacher into a smaller model called the student. By distillating Bert, we obtain a smaller Transformer model that bears a lot of similarities with the original BERT model while being lighter, smaller and faster to run. DistilBERT is thus an interesting option to put large-scaled trained Transformer model into production.
+Distil* is a class of compressed models that started with DistilBERT. DistilBERT stands for Distillated-BERT. DistilBERT is a small, fast, cheap and light Transformer model based on Bert architecture. It has 40% less parameters than `bert-base-uncased`, runs 60% faster while preserving 99% of BERT's performances as measured on the GLUE language understanding benchmark. DistilBERT is trained using knowledge distillation, a technique to compress a large model called the teacher into a smaller model called the student. By distillating Bert, we obtain a smaller Transformer model that bears a lot of similarities with the original BERT model while being lighter, smaller and faster to run. DistilBERT is thus an interesting option to put large-scaled trained Transformer model into production.
We have applied the same method to other Transformer architectures and released the weights:
-- GPT2: on the [WikiText-103](https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/) benchmark, GPT2 reaches a perplexity on the test set of 15.0 compared to 18.5 for **DistilGPT2** (after fine-tuning on the train set).
+- GPT2: on the [WikiText-103](https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/) benchmark, GPT2 reaches a perplexity on the test set of 16.3 compared to 21.1 for **DistilGPT2** (after fine-tuning on the train set).
- RoBERTa: **DistilRoBERTa** reaches 95% of `RoBERTa-base`'s performance on GLUE while being twice faster and 35% smaller.
- German BERT: **German DistilBERT** reaches 99% of `bert-base-german-dbmdz-cased`'s performance on German NER (CoNLL-2003).
- Multilingual BERT: **DistilmBERT** reaches 92% of Multilingual BERT's performance on XNLI while being twice faster and 25% smaller. The model supports 104 languages listed [here](https://github.com/google-research/bert/blob/master/multilingual.md#list-of-languages).
@@ -29,11 +31,11 @@ Here are the results on the dev sets of GLUE:
| Model | Macro-score | CoLA | MNLI | MRPC | QNLI | QQP | RTE | SST-2| STS-B| WNLI |
| :---: | :---: | :---:| :---:| :---:| :---:| :---:| :---:| :---:| :---:| :---: |
-| BERT-base | **77.6** | 48.9 | 84.3 | 88.6 | 89.3 | 89.5 | 71.3 | 91.7 | 91.2 | 43.7 |
-| DistilBERT | **76.8** | 49.1 | 81.8 | 90.2 | 90.2 | 89.2 | 62.9 | 92.7 | 90.7 | 44.4 |
+| BERT-base-uncased | **77.6** | 49.2 | 80.8 | 87.4 | 87.5 | 86.4 | 61.7 | 92.0 | 83.8 | 45.1 |
+| DistilBERT-base-uncased | **76.8** | 43.6 | 79.0 | 87.5 | 85.3 | 84.9 | 59.9 | 90.7 | 81.2 | 56.3 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| RoBERTa-base (reported) | **83.2**/**86.4**2 | 63.6 | 87.6 | 90.2 | 92.8 | 91.9 | 78.7 | 94.8 | 91.2 | 57.73 |
-| DistilRoBERTa1 | **79.0**/**82.3**2 | 59.4 | 83.9 | 86.6 | 90.8 | 89.4 | 67.9 | 92.5 | 88.3 | 52.1 |
+| DistilRoBERTa1 | **79.0**/**82.3**2 | 59.3 | 84.0 | 86.6 | 90.8 | 89.4 | 67.9 | 92.5 | 88.3 | 52.1 |
1 We did not use the MNLI checkpoint for fine-tuning but directy perform transfer learning on the pre-trained DistilRoBERTa.