Fix grammar typos
Browse files
app.py
CHANGED
@@ -47,9 +47,9 @@ _LAST_UPDATED = "Sept 15, 2023"
|
|
47 |
banner_url = "./assets/logo.png"
|
48 |
_BANNER = f'<div style="display: flex; justify-content: space-around;"><img src="{banner_url}" alt="Banner" style="width: 40vw; min-width: 300px; max-width: 600px;"> </div>' # noqa
|
49 |
|
50 |
-
_INTRODUCTION_TEXT = """The 🤗 Nucleotide Transformer Leaderboard aims to track, rank and evaluate DNA foundational models on a set of curated downstream tasks introduced in the huggingface dataset [nucleotide_transformer_downstream_tasks](https://huggingface.co/datasets/InstaDeepAI/nucleotide_transformer_downstream_tasks), with a standardized evaluation
|
51 |
|
52 |
-
This leaderboard has been designed to provide, to the best of our ability, fair and robust comparisons between models. If you have any question or concern regarding our methodology or if you would like another model to appear in
|
53 |
""" # noqa
|
54 |
|
55 |
_METHODS_TEXT = """
|
@@ -59,10 +59,10 @@ This leaderboard uses the downstream tasks benchmark and evaluation methdology d
|
|
59 |
PLease keep in mind that the Enformer has been originally trained in a supervised fashion to solve gene expression tasks. For the sake of benchmarking, we re-used the provided model torso as a pre-trained model for our benchmark, which is not the intended and recommended use of the original paper. Though we think this comparison is interesting to hilight the differences between self-supervised and supervised learning for pre-training and observe that the Enformer is a very competitive baseline even for tasks that differ from gene expression.
|
60 |
\n\n
|
61 |
|
62 |
-
For the sake of clarity the tasks being shown by default in this
|
63 |
\n\n
|
64 |
|
65 |
-
Note also that the performance shown for some methods in
|
66 |
\n\n
|
67 |
""" # noqa
|
68 |
|
|
|
47 |
banner_url = "./assets/logo.png"
|
48 |
_BANNER = f'<div style="display: flex; justify-content: space-around;"><img src="{banner_url}" alt="Banner" style="width: 40vw; min-width: 300px; max-width: 600px;"> </div>' # noqa
|
49 |
|
50 |
+
_INTRODUCTION_TEXT = """The 🤗 Nucleotide Transformer Leaderboard aims to track, rank and evaluate DNA foundational models on a set of curated downstream tasks introduced in the huggingface dataset [nucleotide_transformer_downstream_tasks](https://huggingface.co/datasets/InstaDeepAI/nucleotide_transformer_downstream_tasks), with a standardized evaluation protocol presented in the "ℹ️ Methods" tab.\n\n
|
51 |
|
52 |
+
This leaderboard has been designed to provide, to the best of our ability, fair and robust comparisons between models. If you have any question or concern regarding our methodology or if you would like another model to appear in this leaderboard, please reach out to [email protected] and [email protected]. While we may not be able to take into consideration all requests, the team will always do its best to ensure that benchmark stays as fair, relevant and up-to-date as possible.\n\n
|
53 |
""" # noqa
|
54 |
|
55 |
_METHODS_TEXT = """
|
|
|
59 |
PLease keep in mind that the Enformer has been originally trained in a supervised fashion to solve gene expression tasks. For the sake of benchmarking, we re-used the provided model torso as a pre-trained model for our benchmark, which is not the intended and recommended use of the original paper. Though we think this comparison is interesting to hilight the differences between self-supervised and supervised learning for pre-training and observe that the Enformer is a very competitive baseline even for tasks that differ from gene expression.
|
60 |
\n\n
|
61 |
|
62 |
+
For the sake of clarity the tasks being shown by default in this leaderboard are the human related tasks while the original Nucleotide Transformer paper shows performance over both yeast and human related tasks. To obtain the same results as the one shown in the paper, please check all the tasks boxes above.
|
63 |
\n\n
|
64 |
|
65 |
+
Note also that the performance shown for some methods in this table may differ slightly from the one reported in the HyenaDNA and DNABert papers. This might come from the usage of different train and test splits as well as from our ten-fold systamtic evaluation.
|
66 |
\n\n
|
67 |
""" # noqa
|
68 |
|