Aleph-Alpha/magma

Issues

• OpenSource strategy

  #50 opened 8 months ago by sebastianlutter
  3

• More model structure details expecially parameter numbers of the adapters

  #49 opened 9 months ago by Stephan-XUE
  0

• How are the n-shot VQA examples selected?

  #48 opened a year ago by ys-zong
  0

• ```build_labels``` includes masked image tokens?

  #46 opened a year ago by RealNicolasBourbaki
  3

• How to reproduce the n shot VQA, OKVQA and GQA results from Table -1? No eval script is available

  #45 opened 2 years ago by sanyalsunny111
  2

• Typeguard version in requirements

  #47 opened a year ago by NiclasGoering
  2

• Issue with the dataloader

  #44 opened 2 years ago by sanyalsunny111
  0

• Reproducing results from your paper

  #42 opened 2 years ago by Golovneva
  2

• how did you calculate the bleu score

  #40 opened 2 years ago by TobiasLee
  1

• test

  #35 opened 2 years ago by bashFish
  0

• top_p argument is used like 1-top_p

  #29 opened 2 years ago by nostalgebraist
  1

• AssertionError: Parameter with name: lm.transformer.wte.weight occurs multiple times in optimizer.param_groups. Make sure it only appears once to prevent undefined behaviour

  #32 opened 2 years ago by masoud-monajati
  6

• Torch Size Mismatch

  #34 opened 3 years ago by harshagundala
  1

• Automatic model download doesn't work

  #23 opened 3 years ago by Heath123
  5

• Host checkpoints

  #22 opened 3 years ago by CoEich
  0

• No module named 'magma.transformers'

  #28 opened 3 years ago by Tuxius
  6

• size mismatch for lm.lm_head.weight: copying a param with shape torch.Size([50400, 4096]) from checkpoint, the shape in current model is torch.Size([50258, 4096]).

  #31 opened 3 years ago by UCCME
  1

• Subsequent inference calls produce less good results

  #33 opened 3 years ago by steve-barlow
  1

• Mismatching LM shape between 50400 (pre-trainined pt) and 50258 (gpt-2)

  #27 opened 3 years ago by tsujuifu
  3

• Add LICENSE

  #21 opened 3 years ago by CoEich
  0

• Implement loading checkpoint from url

  #20 opened 3 years ago by CoEich
  0

• Improved inference interface

  #9 opened 3 years ago by CoEich
  1

• Refactor README

  #8 opened 3 years ago by sdtblck
  0

• Update to latest version of transformers

  #5 opened 3 years ago by sdtblck
  1

• Split up model.py into different files

  #13 opened 3 years ago by sdtblck
  1

• clean up train.py

  #17 opened 3 years ago by CoEich
  0

• Clean up model.py / remove codepaths for prefix tuning

  #14 opened 3 years ago by sdtblck
  0

• Cleanup requirements.txt

  #10 opened 3 years ago by sdtblck
  0

• Eval.py: To keep or not to keep?

  #7 opened 3 years ago by sdtblck
  0

• Cleanup `multimodal_fewshot/config.py`

  #4 opened 3 years ago by sdtblck
  0

• Remove dataset builders and old classes in multimodal_fewshot.datasets

  #6 opened 3 years ago by CoEich
  1

• remove `proj_layer_type` Option

  #3 opened 3 years ago by sdtblck
  0

• Clean Up `multimodal_fewshot/adapters.py`

  #2 opened 3 years ago by sdtblck
  0

• Delete Unneeded Configs

  #1 opened 3 years ago by sdtblck
  0