zengxianfeng

Why FSDP is much slower than normal data Parallel?

5

## ❓ Questions and Help ### Before asking: 1. search the issues. 2. search the docs. #### What is your question? I change the `ddp-backend` from `no_c10d` to `fully_sharded` as...

OOM to train translation models with very large multilingual dataset.

4

## ❓ Questions and Help ### Before asking: 1. search the issues. 2. search the docs. #### What is your question? I try to train a multilingual translation model with...

generate.py: error: argument --encoder: invalid choice: 'gpt'

1

I try to use the **MixedCorpus+GptEncoder+LmTarget** model for generate some text like GPT2. And I followed the example scripts like ``` python3 scripts/generate.py --pretrained_model_path models/gpt_model.bin --vocab_path models/google_zh_vocab.txt --input_path story_beginning.txt --output_path...

Can Torch-Pruning be applied to Transformer models ?

3

I try to follow the example to prune the transformer model, but the layer-norm always gets a size mismatch error.

I try to use Post Training Quantization to convert my float32 model to int8 follow the tutorial of quantizing GNMT. I change the model code to a distiller style and...

Is there a way to use dynamic decoding rather than fix output size for the tflite model?

1

The current tflite implementation needs to fix the length of the output sequence,e.g 250. Is there a way to use a really dynamic decodng method?

Do these codes also fit for encoder-decoder transformer?

3

I try to use TUPE in the NMT encoder, but got loss exploding error. Does it need some fix for TUPE to use in NMT? the error is like: ```...

**Describe** I am collecting the corpus used in Deepnet but I can't find where to download the Tatoeba corpus. Is [this](https://github.com/Helsinki-NLP/Tatoeba-Challenge/blob/master/data/README-v2021-08-07.md) what you used in the paper? And I find...

I use the fastBPE to apply bpe to make it faster. But I want to use the origin codes file learned from subword. And I found a clean difference between...

做序列标注 用的OBIE 为什么会出现I O的这种情况 数据里面没有这种情况 这个CRF的实现确定没问题吗？