# tensorflow_modelzoo **Repository Path**: lqy_mycode/tensorflow_modelzoo ## Basic Information - **Project Name**: tensorflow_modelzoo - **Description**: cambriocn tensorflow训练和推理模型集合 - **Primary Language**: Unknown - **License**: Apache-2.0 - **Default Branch**: master - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 33 - **Created**: 2023-07-25 - **Last Updated**: 2024-07-01 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README # TensorFlow ModelZoo ## 介绍 TensorFlow是时下最流行的AI框架，寒武纪对其进行了定制化开发，新增了对寒武纪加速板卡及寒武纪AI软件栈的支持，通常称之为Cambricon TensorFlow。相比于原生TensorFlow，用户基本不用做任何代码改动即可快速地将AI模型迁移至Cambricon TensorFlow上。 针对CV分类、检测、分割、NLP、语音等场景常用的各类经典和前沿的AI模型，本仓库展示了如何对其进行适配，使其可运行在Cambricon TensorFlow上。开发者在进行其他AI应用移植时可参考本仓库。 ## 网络支持列表和链接 CV： | MODELS | FRAMEWORK | Train Mode |Distributed Train| Infer Mode | XLA Support | | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | | [vgg19](tensorflow2/built-in/Classification/common_networks) | TensorFlow2|FP32&& | YES|CNNL|YES| | [resnet50](tensorflow2/built-in/Classification/common_networks) | TensorFlow2|FP32&& | YES|CNNL |YES| | [resnet101](tensorflow2/built-in/Classification/common_networks) | TensorFlow2|FP32&& | YES|CNNL |YES| | [densenet201](tensorflow2/built-in/Classification/common_networks) | TensorFlow2|FP32&& | YES|CNNL |YES| | [swin-transformer](tensorflow2/built-in/Classification/swin-transformer/) | TensorFlow2|FP32&& |YES| CNNL |YES| | [centernet](tensorflow2/built-in/Detection/centernet) | TensorFlow2|FP32&& | YES|CNNL|NO| | [ResNet50](tensorflow/built-in/Classification/common_networks) | TensorFlow1|FP32&& | YES|CNNL |NO| | [ResNet101](tensorflow/built-in/Classification/common_networks) | TensorFlow1|FP32&& | YES|CNNL |NO| | [DenseNet201](tensorflow/built-in/Classification/common_networks) | TensorFlow1|FP32&& | YES|CNNL |NO| | [Vgg19](tensorflow/built-in/Classification/common_networks) | TensorFlow1|FP32&& |YES| CNNL |NO| | [InceptionV3](tensorflow/built-in/Classification/common_networks) | TensorFlow1|FP32&& |YES| CNNL |NO| | [MobilenetV2](tensorflow/built-in/Classification/common_networks) | TensorFlow1|FP32&& |YES| CNNL |NO| | [SSD](tensorflow/built-in/Detection/SSD) | TensorFlow1|FP32&& |YES| CNNL |NO| | [YOLOv3](tensorflow/built-in/Detection/YOLOv3) | TensorFlow1|FP32&& |YES| CNNL |NO| | [retinanet](tensorflow/built-in/Detection/retinanet) | TensorFlow1|FP32&& |YES| CNNL |NO| | [UNet_3D_Medical](tensorflow/built-in/Segmentation/UNet_3D_Medical) | TensorFlow1|FP32&& |YES| CNNL |NO| | [UNet_Industrial](tensorflow/built-in/Segmentation/UNet_Industrial) | TensorFlow1|FP32&& |YES| CNNL |NO| Graph Convolutional Network | MODELS | FRAMEWORK | Train Mode | Distributed Train | Infer Mode | XLA Support | |------------------------------------------------|-------------|------------|-------------------|-------------|-------------| | [GraphSAGE](tensorflow/built-in/GCN/GraphSAGE) | Tensorflow1 | FP32&& | NO | CNNL |NO | NLP: | MODELS | FRAMEWORK | Train Mode |Distributed Train| Infer Mode | XLA Support | | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | | [transformer](tensorflow2/built-in/NaturalLanguageProcessing/transformer) | TensorFlow2|FP32 | YES | CNNL | YES | | [google_bert](tensorflow2/built-in/NaturalLanguageProcessing/google_bert) | TensorFlow2|FP32&& | YES | CNNL | YES | | [tf_models_bert](tensorflow2/built-in/NaturalLanguageProcessing/tf_models_bert) | TensorFlow2|FP32 | YES | CNNL | NO | | [BERT_ngc](tensorflow/built-in/NaturalLanguageProcessing/BERT_ngc) | TensorFlow1|FP32&& | YES | CNNL | NO | | [BERT_CRF](tensorflow/built-in/NaturalLanguageProcessing/bert/bert_crf) | TensorFlow1|FP32&& | YES | CNNL | NO | | [transformer_estimator](tensorflow/built-in/NaturalLanguageProcessing/Transformer/transformer_estimator/) | TensorFlow1|FP32/AMP | YES | CNNL | NO | | [google_bert(tf1)](tensorflow/built-in/NaturalLanguageProcessing/google_bert) | TensorFlow1|FP32&& | YES | CNNL | NO | Recommendation: | MODELS | FRAMEWORK | Train Mode |Distributed Train| Infer Mode | XLA Support | | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | | [DLRM](tensorflow2/built-in/Recommendation/DLRM) | TensorFlow2|FP32&& | YES | CNNL| YES | | [DeepFM](tensorflow/built-in/Recommendation/DeepFM) | TensorFlow1|FP32&& | YES | CNNL| NO | Speech: | MODELS | FRAMEWORK | Train Mode |Distributed Train| Infer Mode | XLA Support | | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | | [Conformer](tensorflow2/built-in/ASR/Conformer) | TensorFlow2|FP32&&|YES| CNNL | No | | [LPCNet](tensorflow2/built-in/TTS/LPCNet) | TensorFlow2|FP32&&|YES| CNNL | No | | [Tacotron2](tensorflow/built-in/TTS/Tacotron-2) | TensorFlow1|FP32&&|YES| CNNL | No | ## issues/wiki/forum 跳转链接 ## contrib 指引和链接 ## Change Log ---------- v1.5.1 2023年7月12日 ---------- - 修复了dlrm的一个 rank 的bug，更新了change log. ---------- v1.5.0 2023年7月12日 ---------- - 修复了一些bug，tf modelzoo v1.5.0与tf v1.19.0版本相对应 ---------- v1.4 2023年5月17日 ---------- - 向 tensorflow2 目录内添加了 tf_models_bert 网络 ---------- 2023年5月9日 ---------- - 支持网络列表删除了tensorflow1中的Resnet18,Alexnet,Vgg16,Inceptionv2,Resnet50-1.5，删除了tensorflow2中的resnet50_tf_model_official_v2.8.0,vgg16,resnet18。 - 添加了网络是否支持XLA模式的说明。 ---------- v1.3 2023年3月31日 ---------- 将tensorflow2/built-in 内使用 TFMM 进行推理的网络改为使用 CNNL 进行推理，涉及到的网络如下： [vgg16](tensorflow2/built-in/Classification/common_networks) [vgg19](tensorflow2/built-in/Classification/common_networks) [resnet18](tensorflow2/built-in/Classification/common_networks) [resnet50](tensorflow2/built-in/Classification/common_networks) [resnet101](tensorflow2/built-in/Classification/common_networks) [densenet201](tensorflow2/built-in/Classification/common_networks) [centernet](tensorflow2/built-in/Detection/centernet) [transformer](tensorflow2/built-in/NaturalLanguageProcessing/transformer) [google_bert](tensorflow2/built-in/NaturalLanguageProcessing/google_bert) ## LICENSE TensorFlow ModelZoo 的 License 具体内容请参见[LICENSE](LICENSE)文件。 ## 免责声明 TensorFlow ModelZoo 仅提供公共数据集以及预训练模型的下载链接，公共数据集及预训练模型并不属于 TensorFlow ModelZoo ，TensorFlow ModelZoo 也不对其质量或维护承担责任。请您在使用公共数据集和预训练模型的过程中，确保符合其对应的使用许可。 如果您不希望您的数据集或模型公布在 TensorFlow ModelZoo上，或者您希望更新 TensorFlow ModelZoo中属于您的数据集或模型，请您通过 Gitee 中提交 issue，您也可以联系ecosystem@cambricon.com告知我们。