TurboTransformers
TurboTransformers copied to clipboard
a fast and user-friendly runtime for transformer inference (Bert, Albert, GPT2, Decoders, etc) on CPU and GPU.
你好,请问用tensorflow加载bert,是先将bert权重转成npz格式,再用`from_npz `加载么?`from_torch`和`from_pretrained`应该都都是加载的pytorch模型,不知道我理解得对不对?另外,请问gpt2现在还只能用onnx加速么?
你好,想请教一个关于 Tensor 的问题: 假设我现在有一个 {32,128,768} 维度的 Tensor,现在想去掉第二个维度的头尾向量,将 Tensor 的维度变为 {32,126,768},请问有什么好的解决办法吗?
是不是仅支持README中提到的几个模型以下几个模型?假设基于pytorch搭建一个transformer模型,能用turbo transformer加速吗? BERT [Python] [C++] ALBERT [Python] Roberta [Python] Transformer Decoder [Python] GPT2 [Python]
Hi,feifeibear 我仿照您提供的例子(bert_for_sequence_classification_example.py)写了个用于命名实体识别的测试程序,运行下来发现Turbo提速并不明显,请问这中间有什么潜在的注意事项吗? 
大家好, 我在hugging face上面下载了bert-base-uncased模型来测试bert_example.py脚本,操作完全按照TurboTransformers/example/python/README.md这个里面来操作,不过因为本地编译有问题,所以我这里用的是作者提供的编译好的镜像(docker pull thufeifeibear/turbo_transformers_cpu:latest),结果报以下错误: **Traceback (most recent call last): File "", line 3, in File "/opt/miniconda3/lib/python3.7/site-packages/turbo_transformers/layers/modeling_bert.py", line 495, in __call__ outputs = self.onnxmodel.run(inputs=data) AttributeError: 'BertModelNoPooler' object has no attribute...
你好: 我下载最新的turbotransformers程序编译,使用命令sh tools/build_docker_cpu.sh 这一步就报了很多库冲突,这个问题该怎么解决呢,一部分报错信息如下 > Package jupyterlab conflicts for: > _ipyw_jlab_nb_ext_conf -> jupyterlab > anaconda==2021.05 -> jupyterlab==3.0.14=pyhd3eb1b0_1 > > Package perl conflicts for: > git -> perl[version='5.*|>=5.26.2,=5.26.2,=1.3.7,=1.18.3,=1.3.7,=4.3'] -> conda-env[version='>=2.6'] >...
The benchmark testing script just shows the QPS or latency comparison without accuracy testing. Does turbo have the method for accuracy testing or some suggestion to do this? Thanks.
Below is a rough code to explain what I did. ```python import torch.multiprocessing as mp def inference(config): data_loader = get_loader(config) while True: for step in range(128): dec_outs, _ = turbo_decoder(current_pred,...
It would be good if you add java support with JavaCPP. So this library can work on IOS , OSX , Android , Linux , Windows ?