-
```
ometimes when trying to re-print a label for a second time I get the following
error:
FireFox Fire Bug:
applet.getVersion is not a function
Line 45
INFO: ===== JAVASCRIPT LISTENER THREAD STOPP…
-
您好,我在复现您代码的时候,出现错误:
RuntimeError: CUDA error: device-side assert triggered
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
错误地点是pooled_logits = logits[torch.arange(batch_size, d…
-
# 1. 기존 Pretrained Model 문제점 (?) 아쉬운 점(?)
기존의 Pretrained Model을 불러와서 쓰면 항상 고정된 classifier에 label의 개수만 바꿔주는 형식으로 동작.
ex) `config.num_labels=30`
Transformer에서 AutoModelForSequenceClassification을 살펴보시…
-
### System Info
transformers>=4.43.2
accelerate==0.33.0
GPUs: 2 x A100
### Who can help?
@ArthurZucker @muellerzr @SunMarc
### Information
- [ ] The official example scripts
- [ ] My own modif…
-
I have no idea of this line about sampling from softmax. Can you give me a help?
```
sel = tf.one_hot(tf.argmax(logit_probs - tf.log(-tf.log(tf.random_uniform(logit_probs.get_shape(), minval=1e-5, m…
-
This can be done with logit filters on the first loop, similar to detecting language. However, this cannot be used when we are using a prefill prompt (i.e. forced decoder tokens) so that will need spe…
-
```
File C:\.conda\envs\gettsim-application\lib\site-packages\pylogit\__init__.py:10
2 """
3 Created on Mon Mar 14 15:33:07 2016
4
5 @author: timothyb0912
6 @modul…
-
Is there any possibility to get start and end timestamp both for every token, currently we are getting only start time of every token.
A scenario that we face in general is, we have multiple speech…
-
line 137: predict = []
line 145: predict += logits[0]
会报错:operands could not be broadcast together with shapes (0,) (n,)
(n是label个数)
原因应该是在循环中空数组和预测值logits的第0维相加出现了维度问题。
我改成了 line 137: predic…
-
Thanks for this project !
This project is awesome. However I have an issue that occur really often and I can not figure out how to solve it :/
I am testing the repository in multi_frame using your…