patrickvonplaten
Follow
Create your own GitHub profile
Sign up for your own profile on GitHub, the best place to host code, manage projects, and build software alongside 50 million developers.
Sign up
Popular repositories
1,079 contributions in the last year
Contribution activity
July 2020
Created a pull request in huggingface/transformers that received 7 comments
[Generation] better error message
If cur_len of input context is as long or longer than max_length a nice error message should be shown.
+8
−0
•
7
comments
- [WIP - TF Longformer] Add TF Longformer
- [Don't merge - Bert2Bert] Add training scripts and slight changes to Trainer
- [AutoModels] Fix config params handling of all PT and TF AutoModels
- [WIP - don't merge][TF generate] Make tf generate compatible with tf.function
- [Longformer] fix longformer global attention output
- Update benchmark notebook
- [Benchmark] Add benchmarks for TF Training
- [WIP - Don't merge][Reformer] - Cache hidden states and buckets to speed up inference
- [Reformer] Adapt Reformer MaskedLM Attn mask
- [Don't merge] Reformer Trax Integration Tests
- Add Reformer MLM notebook
- [Reformer] Add QA head to reformer model
- [Reformer] Add Masked LM Reformer
- Finalize Reformer blog with fixed Latex
- Reformer - fix more underscore
- fix reformer underscore
- fix reformer formulas
- fix latex in reformer blog
- Never doing words in latex in markdown again
- try out new word regex
- Last time words fixing in generate
- fix words
- Fix words generate
- fix generate blog in markdown
- Fix how to generate
- Add reformer notebook
- [cleanup] T5 test, warnings
- [WIP - Don't merge][Reformer] - Cache hidden states and buckets to speed up inference
- doc: fix apparent copy-paste error in docstring
- Cleanup bart caching logic
- [fix] mbart_en_ro_generate test now identical to fairseq
- [Reformer classification head] Implement the reformer model classification head for text classification
- [Longformer] fix longformer global attention output
- FlaubertForTokenClassification
- Fixed use of memories in XLNet (caching for language generation + warning when loading improper memoryless model)
- Improvements to PretrainedConfig documentation
- Fixed TextGenerationPipeline on torch + GPU
- [Benchmark] Add benchmarks for TF Training
- Add mbart-large-cc25, support translation finetuning
- [Reformer] Adapt Reformer MaskedLM Attn mask
- Added data collator for permutation (XLNet) language modeling and related calls
- The `add_space_before_punct_symbol` is only for TransfoXL
- Refactor generation sampling parameters (e.g. top k, temperature) into "Sampling" classes
- Fix saved model creation
- [Bart] enable test_torchscript, update test_tie_weights
- QA pipeline BART compatible
- Change model outputs types to self-document outputs
- [Almost all TF models] TF clean up: add missing CLM / MLM loss; fix T5 naming and keras compile
- Add Reformer MLM notebook
- Gradient checkpointing BERT & ALBERT poc
- MarianTokenizer.prepare_translation_batch uses new tokenizer API
- Some pull request reviews not shown.