Closed yongyi-wu closed 3 years ago
Merging #1506 (eddda7d) into master (f44ec96) will decrease coverage by
0.12%
. The diff coverage is95.68%
.
@@ Coverage Diff @@
## master #1506 +/- ##
==========================================
- Coverage 86.35% 86.23% -0.13%
==========================================
Files 54 55 +1
Lines 7388 7502 +114
==========================================
+ Hits 6380 6469 +89
- Misses 1008 1033 +25
Impacted Files | Coverage Δ | |
---|---|---|
src/gluonnlp/attention_cell.py | 88.39% <ø> (ø) |
|
src/gluonnlp/models/mt5.py | 95.49% <95.49%> (ø) |
|
src/gluonnlp/models/__init__.py | 100.00% <100.00%> (ø) |
|
src/gluonnlp/models/t5.py | 93.67% <100.00%> (+0.01%) |
:arrow_up: |
src/gluonnlp/models/transformer.py | 98.52% <100.00%> (+<0.01%) |
:arrow_up: |
src/gluonnlp/data/loading.py | 75.84% <0.00%> (-7.55%) |
:arrow_down: |
Continue to review full report at Codecov.
Legend - Click here to learn more
Δ = absolute <relative> (impact)
,ø = not affected
,? = missing data
Powered by Codecov. Last update f44ec96...eddda7d. Read the comment docs.
The documentation website for preview: http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR1506/ec54d05bd8331c5ee50b0ff169818952dd79846c/index.html
The documentation website for preview: http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR1506/eddda7db3c822ff773658c613df8ad1128b41d87/index.html
Description
This PR introduces mT5 models and conversion scripts to GluonNLP codebase. The models, tokenizer, conversion scripts, and test cases reuse most of T5's, so the correctness largely depends on the implementation of T5's components. Besides, docstrings for all T5 building blocks have been added.
Checklist
Essentials
Changes
MT5Model
,MT5Inference
, andMT5Tokenizer
cc @dmlc/gluon-nlp-team