fairseq

fairseq copied to clipboard

Can the fairseq-13b model be used commercially? Which license applies?

2

Hi, Regarding the models listed here https://github.com/pytorch/fairseq/tree/main/examples/moe_lm the model card and NOTE file included with the models say "Models are intended for research use only". Does this mean that they...

timohear

AssertionError: lexicon free decoding can only be done with a unit language model

7

## ❓ Questions and Help ### Before asking: 1. search the issues. 2. search the docs. #### What is your question? When I evaluate a CTC model on wav2cec2.0 according...

quhonglin

NLLB inference time model loading fails due to inconsistent vocabulary size

8

Hi, I downloaded the dictionary and 600M NLLB-200-Distilled checkpoint. I failed to load model weights from the checkpoint due to inconsistent vocabulary size. The dictionary has 255997 tokens and the...

pluiez

Hi, I'm trying to train wav2vec2 model with my custom hindi data. But the issue I'm facing is I'm not sure what are the characters of Hindi dict file will...

MrityunjoyS

Lightning App example for nllb (coming soon)

3

@myleott @alexeib amazing work once again! Would love to collaborate with you guys to build a Lightning App (a new class available in Lightning) so people can your work as...

williamFalcon

Do any fairseq CLI's support forced decoding?

2

## ❓ Questions and Help #### What is your question? Is there some code path to enable forced decoding from the CLI? #### Code It seems like adding encoded inputs...

erip

Hi, this PR adds an check if the `target_model` module really exists in the encoder of a data2vec model. With newer pretrained data2vec models, this check is really important because...

stefan-it

TypeError: forward() missing 1 required positional argument: 'prev_output_tokens'

5

## 🐛 Bug ### To Reproduce Steps to reproduce the behavior (**always include the command you ran**): 1. Run cmd fairseq-train $DATASET_DIR --arch $BART --restore-file $PRETRAINED_MODEL \ --save-dir $SAVE_MODEL_DIR --tensorboard-logdir...

ghost

fp16 bug about grad_norm computation (resulting in min loss scale reached)

3

## 🐛 Bug [FloatingPointError: Minimum loss scale reached (0.0001). ](https://github.com/facebookresearch/fairseq/issues/1529#issuecomment-567955547) I met this assertation often when do training in fp16. I found it may be avoided by modifying the way...

flycser

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...

SefaZeng