patrickvonplaten Follow

patrickvonplaten Follow

Patrick von Platen patrickvonplaten

Follow

122 followers · 24 following · 61

Sign in to view email

Highlights

Arctic Code Vault Contributor
Pro

Organizations

Popular repositories

notebooks

Some notebooks for NLP

Jupyter Notebook 21 7
AdvertisementDetection

Python 2
TRexGameRL

JavaScript 2
course_work

TeX 1 1
AdvancedAutomaticSpeechRecognition

Python 1
DM_Stat_avance

TeX

1,335 contributions in the last year

Contribution activity

October 1, 2020

patrickvonplaten has no activity yet for this period.

September 2020

patrickvonplaten/datasets-1 Python Sep 14

Created a pull request in huggingface/transformers that received 11 comments

[Longformer, Bert, Roberta, ...] Fix multi gpu training

Fixes #6256. Issue #6256 shows that distributed training is not possible when the model has layers that are not used at all. Bert, Roberta and Lon…

+179 −49 • 11 comments

[Seq2Seq] Fix a couple of bugs and clean examples Sep 30
[RAG] Model cards - clean cards Sep 28
[Rag] fix rag retriever save_pretrained method Sep 25
[Rag] Fix wrong usage of `num_beams` and `bos_token_id` in Rag Sequence generation Sep 25
[RAG] Add missing doc and attention_mask to rag Sep 25
[RAG] Add `attention_mask` to RAG generate Sep 24
[RAG] PR to save status of previous RAG code Sep 17
[EncoderDecoderModel] fix indentation error Sep 14
[WIP RAG] Finalize RAG parallel Sep 14
[BertGeneration] Clean naming Sep 11
[BertGeneration, Docs] Fix another old name in docs Sep 10
[BertGeneration] Correct Doc Title Sep 10
[Longformer] Fix longformer documentation Sep 8
[WIP] Refactoring the generate() function Sep 4
[LXMERT] Fix tests on gpu Sep 4
Torchscript benchmark measure Sep 2
[Docs, Examples] Fix QA example for PT Sep 1
[Electra] fix warning for position ids Sep 1
Create README.md Sep 1
[EncoderDecoder] Add xlm-roberta to encoder decoder Sep 1
[WIP, TF] replace keras dense by keras.layers.DenseEinsum Sep 1

[WIP] Facebook RAG Sep 14

Update pytorch_block_sparse.md Sep 3

[Seq2Seq] Fix a couple of bugs and clean examples Sep 30
Add DeBERTa model Sep 29
Adding gradient checkpointing to GPT2 Sep 29
SqueezeBERT architecture Sep 29
Custom TF weights loading Sep 28
Make T5 compatible with ONNX Sep 28
[WIP] ProphetNet Sep 28
[RAG] Clean Rag readme in examples Sep 28
[T5] allow config.decoder_layers to control decoder size Sep 27
Replaced torch.load for loading the pretrained vocab of TransformerXL tokenizer to pickle.load Sep 25
[RAG] Remove dependency on `examples/seq2seq` from rag Sep 25
Enable pegasus fp16 by clamping large activations Sep 25
Remove unhelpful bart warning Sep 25
[RAG] Fix retrieval offset in RAG's HfIndex and better integration tests Sep 25
Make PyTorch model files independent from each other Sep 24
Clean RAG docs and template docs Sep 24
[Benchmarks] Change all args to from `no_...` to their positive form Sep 23
[Longformer, Bert, Roberta, ...] Fix multi gpu training Sep 22
[Bug fix] Fixed target_mapping preparation for XLNet (Pytorch) Sep 20
Add tests and fix various bugs in ModelOutput Sep 11
fix deprecation warnings Sep 10
Add "Leveraging Pretrained Checkpoints for Generation" Seq2Seq models. Sep 9
[from_pretrained] Allow tokenizer_type ≠ model_type Sep 8
[generation] consistently add eos tokens Sep 7
[gen utils] missing else case Sep 7
Some pull request reviews not shown.

Created an issue in huggingface/transformers that received 1 comment

Missing keys when loading weights in TF are not useful

This concerns all TF models If one loads weights of a tensorflow model these lines are run: transformers/src/transformers/modeling_tf_utils.py …

1 comment

You can’t perform that action at this time.