lonelydancer

Results 16 issues of lonelydancer

I'm using a bart model for text_generation.i modified https://github.com/lonelydancer/algorithm/blob/master/ls_bart2.py https://github.com/lonelydancer/algorithm/blob/master/hf_bart_export.py the generated text is wrong. ![image](https://user-images.githubusercontent.com/548443/148677299-1450fc72-8f7d-40a5-8587-ca426bf21ac0.png)

in examples/inference/python/export/export_bart.py, why do u do ***Fix encoder layer {} LayerNorm scale and bias*** ***Fix decoder layer {} LayerNorm scale and bias*** it seems to assign the weight of layer(i+1)...

### Describe the bug run the example in Get Started in 60 Seconds ![image](https://user-images.githubusercontent.com/548443/127103356-984ec2c8-2c09-4766-accb-c0c65186c3ce.png) ### Context - **OS** Ubuntu18.04 - **Hardware** Tesla 80k, cuda 10.1,cudnn7.0 - matchzoo 2.2.0, tensorflow2.2.0, keras2.3.0...

bug

RT 据我了解keras&paddle都可以用generator来解决。不知道pytorch用什么方法解决。

Hi, Why ds = dsv + diff_s what is the difference between diff_s and ds? I'am confused. thank you!

1)我看paper中的NSPModel,“To select the most appropriate responses generated by the fine-grained generation model, the evaluation model is trained to estimate the coherence of the responses.” 理解为用stage 2.1生成的候选 + label 做分类model 而代码中的...

你好,我在一些数据上重训nsp model,发现mask策略会使tgt_label采样为空。 具体在nsp_reader.py 的_pad_batch_records函数中 batch_mask_token_ids, tgt_label, tgt_pos, label_pos = mask( batch_tokens=batch_token_ids, vocab_size=self.vocab_size, bos_id=self.bos_id, eos_id=self.eos_id, mask_id=self.mask_id, sent_b_starts=batch_tgt_start_idx, labels=batch_label, is_unidirectional=False) 而mask策略,多次采样有时候prob 均> 0.15 ,导致mask_label、mask_pos都为空。 我在这块多次采样直到非空,暂时解决了这个问题。

![image](https://user-images.githubusercontent.com/548443/127103641-8a56ad30-b5cb-406a-bc45-4a2aed7c7089.png) 在40%-80%之间跳。 是实现有bug?

我改了一个版本,主要是输入的时候一个query下有一个正例和4个负例,用代码中的网络结构,发现预测的时候无论样本是什么样的,第一个样本永远是正例。 请问跟网络结构实现有关系,导致第一个网络的权重特别大?应该共享一个网络会不会变好?