Qing

ControlNet is so great! I developed a tiny little diffusion drawing app to make it easier to use the scribble model. It is free and open source, and you can...

**Is your feature request related to a problem? Please describe.** I am using LoRA + deepspeed stage3 to train the Bloom model. One of the advantages of using LoRA is...

### Is your feature request related to a problem? Please describe. _No response_ ### Solutions 请问是否有关于 SFT 训练数据量、数据类型分布的技术报告？ ### Additional context _No response_

Hi, thank you for your outstanding work. I am adding an [outpainting function](https://github.com/Sanster/lama-cleaner) in my open-source project and have a question to ask: How can I modify the `soften_mask` function...

网络效率测试

3

计算网络的计算量，参考 https://stackoverflow.com/questions/45085938/tensorflow-is-there-a-way-to-measure-flops-for-a-model 运行对应的 net 文件获得 profile 信息，如 `python3 squeeze_net.py` |网络|参数个数(M)|计算量 (MFLOP)|Time (s, CPU batch_size=128)|Time (s, GPU 1080 Ti batch_size=128)| |---|--------|-----------|-----|---| |Paper Raw| 5.55 |3142.43|2.6|0.322| |Simple Net|4.96|2708.19|2.7|0.279| |Squeeze Net|0.72|863.12|3.7|0.552| |ResNet|3.49|2208.28|6.5|0.796|

参考

1

- cat: https://github.com/microsoft/nni/blob/539a7cd7d8c7abc0cb348c6eefaf8b4e01ff18d0/nni/compression/pytorch/speedup/compressor.py#L97 - group conv: https://github.com/666DZY666/micronet/blob/master/micronet/compression/pruning/gc_prune.py

ICDAR

14

**ICDAR13** - Train step: 80k - lr: 0.00001 | Net | Dataset | Recall | Precision | Hmean | |-------|----------|---------|-------------|------------| | Origin CTPN |ICDAR13 + ?|73.72% | 92.77% | 82.15%|...

Where can I find paper

2

Thank you for open source your research code. Where can I find the paper? Thanks

Question about `--per-sequence-loss`

1

In `generate_dataset.py`, there is a `--per-sequence-loss` arg, which used in `conversation_template.py`. This parameter further adjusts the weights based on the length of each response. https://github.com/imoneoi/openchat/blob/30da91b20f11bf5aa268e84b6f5587caa37f510f/ochat/config/conversation_template.py#L104 I would like to know,...

关于模型评估过程中的标准化操作

2

在新版公式识别的 blog 中提到了计算 CER 前会对标注和模型结果进行标准化，目前我也在做这方面的评估，采用了以下的标准化操作，但是不确定这些操作是否足够，想请教一下还有没有其它的标准化操作能够提高测试指标的准确程度？ - 去除空格符号 `[r"\ ", r"\quad", r"\,", r"\;", r"\:", r"\!", " "]` - 统一 environment： 例如 "equation", "align", "align*", "gathered", "$$", "\[", "\]" 等