kssteven418/I-BERT

Issues

• python run.py --arch roberta_base --task STS-B

  #33 opened 2 months ago by Arthur-Ling
  0

• Can not inference the quantilized model in my device by int8

  #15 opened 3 years ago by deepfind
  1

• About CUDA out of memory

  #32 opened 3 months ago by Ai-ZL
  0

• Task name references in strings are wrong

  #29 opened a year ago by AmosHason
  1

• Cannot perform quantization-aware-finetuning due to NaN values

  #31 opened a year ago by AmosHason
  0

• Where can I find the integer-sqrt kernel ?

  #26 opened 2 years ago by Alex-Songs
  0

• 0

  #25 opened 2 years ago by Arouniversal
  0

• About scaling_factor

  #24 opened 2 years ago by DOHA-HWANG
  0

• Pre-trained weights for specific tasks

  #23 opened 2 years ago by roymiles
  0

• IBert problems of quant_model=true

  #18 opened 3 years ago by CSuperlei
  1

• Storing both float32 and int parameters

  #22 opened 2 years ago by huu4ontocord
  0

• Latency 20x with quant_mode = true

  #21 opened 2 years ago by LiamPKU
  1

• Arguments in run.py

  #20 opened 2 years ago by ZhangYunchenY
  0

• Wrong script of downloading GLUE datasets

  #19 opened 2 years ago by Alexiazzf
  0

• Another setting for quantization

  #17 opened 3 years ago by gksruf
  0

• How can we change the quantization settings?

  #16 opened 3 years ago by kentaroy47
  0

• why is Integer-only finetuning is much more slower than fp32 finetune

  #14 opened 3 years ago by renmada
  0

• Is ibert-roberta-base on huggingface model hub the same as roberta-base

  #13 opened 3 years ago by renmada
  0

• How is the scaling factor S implemented with integer?

  #12 opened 3 years ago by deJQK
  0

• 3rd order polynomial approximation to GeLU

  #11 opened 3 years ago by CaoZhongZ
  0

• Bugs in the code

  #10 opened 3 years ago by hsiehjackson
  0

• rationale considering in using floor or round

  #9 opened 3 years ago by CaoZhongZ
  1

• (huggingface) The output of IBERT is float. Am I doing wrong?

  #8 opened 3 years ago by kyoungrok0517
  0

• quantize other roberta model

  #7 opened 3 years ago by longyueling
  0

• Quantization on trained model

  #6 opened 3 years ago by shon-otmazgin
  10

• Missing deployment part on TensorRt

  #2 opened 3 years ago by pommedeterresautee
  3

• Training in mixed precision

  #3 opened 3 years ago by bdalal
  3

• Why use 22 bit quantized activations for some layer norms (except in Embeddings)?

  #5 opened 3 years ago by bdalal
  2

• Possible bug in IntSoftmax

  #4 opened 3 years ago by bdalal
  0

• Can use the CPU in the inference state?

  #1 opened 3 years ago by luoling1993
  1