kssteven418 I-BERT issues

kssteven418 / I-BERT

[ICML'21 Oral] I-BERT: Integer-only BERT Quantization

https://arxiv.org/abs/2101.01321

MIT License

226 stars 32 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

error when runing download_glue_data.py

#34 amosrohe opened 3 months ago
0
python run.py --arch roberta_base --task STS-B

#33 Arthur-Ling opened 4 months ago
0
About CUDA out of memory

#32 Ai-ZL opened 6 months ago
0
Cannot perform quantization-aware-finetuning due to NaN values

#31 AmosHason opened 1 year ago
0
Fix NaN values in IntLayerNorm

#30 AmosHason opened 1 year ago
1
Task name references in strings are wrong

#29 AmosHason opened 1 year ago
1
Fix STS-B name reference

#28 AmosHason opened 1 year ago
1
Fix task name references

#27 AmosHason opened 1 year ago
1
Where can I find the integer-sqrt kernel ?

#26 Alex-Songs opened 2 years ago
0
0

#25 Arouniversal opened 2 years ago
0
About scaling_factor

#24 DOHA-HWANG opened 2 years ago
0
Pre-trained weights for specific tasks

#23 roymiles opened 2 years ago
0
Storing both float32 and int parameters

#22 huu4ontocord opened 2 years ago
0
Latency 20x with quant_mode = true

#21 LiamPKU opened 2 years ago
1
Arguments in run.py

#20 ZhangYunchenY opened 2 years ago
0
Wrong script of downloading GLUE datasets

#19 Alexiazzf opened 2 years ago
0
IBert problems of quant_model=true

#18 CSuperlei opened 2 years ago
1
Another setting for quantization

#17 gksruf opened 2 years ago
0
How can we change the quantization settings?

#16 kentaroy47 opened 3 years ago
0
Can not inference the quantilized model in my device by int8

#15 deepfind opened 3 years ago
1
why is Integer-only finetuning is much more slower than fp32 finetune

#14 renmada closed 3 years ago
0
Is ibert-roberta-base on huggingface model hub the same as roberta-base

#13 renmada closed 3 years ago
0
How is the scaling factor S implemented with integer?

#12 deJQK opened 3 years ago
0
3rd order polynomial approximation to GeLU

#11 CaoZhongZ closed 3 years ago
0
Bugs in the code

#10 hsiehjackson opened 3 years ago
0
rationale considering in using floor or round

#9 CaoZhongZ opened 3 years ago
1
(huggingface) The output of IBERT is float. Am I doing wrong?

#8 kyoungrok0517 closed 3 years ago
0
quantize other roberta model

#7 longyueling opened 3 years ago
0
Quantization on trained model

#6 shon-otmazgin opened 3 years ago
10
Why use 22 bit quantized activations for some layer norms (except in Embeddings)?

#5 bdalal closed 3 years ago
2
Possible bug in IntSoftmax

#4 bdalal opened 3 years ago
0
Training in mixed precision

#3 bdalal closed 3 years ago
3
Missing deployment part on TensorRt

#2 pommedeterresautee opened 3 years ago
3
Can use the CPU in the inference state?

#1 luoling1993 opened 3 years ago
1