Skip to content
This repository has been archived by the owner on Jan 15, 2024. It is now read-only.

[WIP] Update xlnet-large glue results #1104

Merged
merged 11 commits into from
Jan 14, 2020
Merged

[WIP] Update xlnet-large glue results #1104

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
41e0830
Merge pull request #1 from dmlc/master
zburning Oct 25, 2019
a10c89a
Update transformer.py
zburning Oct 25, 2019
23be6c6
Update transformer.py
zburning Oct 25, 2019
090a6cd
Update transformer.py
zburning Oct 25, 2019
358169f
Update transformer.py
zburning Oct 26, 2019
f49b536
Merge pull request #2 from dmlc/master
zburning Oct 29, 2019
278340c
Update transformer.py
zburning Oct 29, 2019
101ce13
Merge pull request #3 from dmlc/master
zburning Oct 31, 2019
2fe5744
Merge remote-tracking branch 'upstream/master'
Jan 10, 2020
aeed23e
small changes and xlnet large results
Jan 10, 2020
aba6feb
fix
Jan 13, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 21 additions & 1 deletion scripts/language_model/index.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -236,13 +236,33 @@ Results using `xlnet_12_768_12`:
+-----------------+---------------------+-----------------------+--------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+
| QQP |Accuracy |90 |`log <https://github.com/dmlc/web-data/tree/master/gluonnlp/logs/language_model/xlnet_l12_h768_a12_finetuned_QQP.log>`__ |`command <https://github.com/dmlc/web-data/tree/master/gluonnlp/logs/language_model/xlnet_l12_h768_a12_finetuned_QQP.sh>`__ |
+-----------------+---------------------+-----------------------+--------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+
| MNLI |Accuracy |87/86 |`log <https://github.com/dmlc/web-data/tree/master/gluonnlp/logs/language_model/xlnet_l12_h768_a12_finetuned_MNLI.log>`__ |`command <https://github.com/dmlc/web-data/tree/master/gluonnlp/logs/language_model/xlnet_l12_h768_a12_finetuned_MNLI.sh>`__ |
| MNLI |Accuracy(m/mm) |87/86 |`log <https://github.com/dmlc/web-data/tree/master/gluonnlp/logs/language_model/xlnet_l12_h768_a12_finetuned_MNLI.log>`__ |`command <https://github.com/dmlc/web-data/tree/master/gluonnlp/logs/language_model/xlnet_l12_h768_a12_finetuned_MNLI.sh>`__ |
+-----------------+---------------------+-----------------------+--------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+
| QNLI |Accuracy |88 |`log <https://github.com/dmlc/web-data/tree/master/gluonnlp/logs/language_model/xlnet_l12_h768_a12_finetuned_QNLI.log>`__ |`command <https://github.com/dmlc/web-data/tree/master/gluonnlp/logs/language_model/xlnet_l12_h768_a12_finetuned_QNLI.sh>`__ |
+-----------------+---------------------+-----------------------+--------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+
| RTE |Accuracy |74 |`log <https://github.com/dmlc/web-data/tree/master/gluonnlp/logs/language_model/xlnet_l12_h768_a12_finetuned_RTE.log>`__ |`command <https://github.com/dmlc/web-data/tree/master/gluonnlp/logs/language_model/xlnet_l12_h768_a12_finetuned_RTE.sh>`__ |
+-----------------+---------------------+-----------------------+--------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+

Results using `xlnet_24_1024_16`:
We followed the hyperparameters reported by the paper authors.

+-----------------+---------------------+-----------------------+--------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+
|Task Name |Metrics |Results on Dev Set |log |command |
+=================+=====================+=======================+============================================================================================================================================+=================================================================================================================================================================+
| CoLA |Matthew Corr. |67 |`log <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_CoLA.log>`__ |`command <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_CoLA.sh>`__ |
+-----------------+---------------------+-----------------------+--------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+
| SST-2 |Accuracy |94 |`log <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_SST.log>`__ |`command <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_SST.sh>`__ |
+-----------------+---------------------+-----------------------+--------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+
| MRPC |Accuracy/F1 |90.2/93 |`log <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_MRPC.log>`__ |`command <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_MRPC.sh>`__ |
+-----------------+---------------------+-----------------------+--------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+
| STS-B |Pearson Corr. |91.37 |`log <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_STS-B.log>`__ |`command <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_STS-B.sh>`__ |
+-----------------+---------------------+-----------------------+--------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+
| QQP |Accuracy |91.94 |`log <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_QQP.log>`__ |`command <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_QQP.sh>`__ |
+-----------------+---------------------+-----------------------+--------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+
| MNLI |Accuracy(m/mm) |89.93/89.91 |`log <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_MNLI.log>`__ |`command <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_MNLI.sh>`__ |
+-----------------+---------------------+-----------------------+--------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+
| RTE |Accuracy |84.12 |`log <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_RTE.log>`__ |`command <https://github.com/dmlc/web-data/blob/master/gluonnlp/logs/language_model/xlnet_l24_h1024_a16_finetuned_RTE.sh>`__ |
+-----------------+---------------------+-----------------------+--------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+



2 changes: 1 addition & 1 deletion scripts/language_model/model/XLNet_classifier.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,7 @@ def __init__(self, xl, units=768, num_classes=2, dropout=0.0,
self.classifier = nn.HybridSequential(prefix=prefix)
if dropout:
self.classifier.add(nn.Dropout(rate=dropout))
self.classifier.add(nn.Dense(units=num_classes))
self.classifier.add(nn.Dense(units=num_classes, flatten=False))
self.pooler = nn.Dense(units=units, flatten=False, activation='tanh', prefix=prefix)

def __call__(self, inputs, token_types, valid_length=None, mems=None):
Expand Down
Loading