tensorflow 2.16.1 can not use save and load

关注

原创最新推荐文章于 2026-08-18 09:25:40 发布 994 阅读 AI 写同款· GEO 优化 ›

从Notebook到生产：机器学习模型服务化落地实战机器学习模型服务化是将训练完成的模型稳定、高效、可观测地部署为线上API的关键环节。其核心在于解决推理延迟、特征一致性、模型热更新与生产可观测性等工程挑战。Triton Inference Server凭借多框架支持、动态批处理和GPU优化，成为高性能模型服务的事实标准；Feature Store则通过统一特征定义与编译执行，消除训练-服务偏差。该过程远超简单封装API，需融合MLOps规范、SRE实践与闭环反馈机制，广泛应用于推荐系统、风控引擎与IoT预测性维护等高要求场景。阅读详情

already use model.save(). not use the load weight. the question is : self.b1 = BatchNormalization(), in my code, the BatchNormalization(), do not take any parameter, why Layer 'batch_normalization_20' expected 4 variables?

Sent by you: already use model.save(). not use the load weight. the question is : self.b1 = BatchNormalization(), in my code, the BatchNormalization(), do not take any parameter, why Layer 'batch_normalization_20' expected 4 variables?

change to 2.15.0 , it is ok. but a lot of change on

model_res_net.compile(optimizer=optimizer,
                      # loss=[tf.keras.losses.CategoricalCrossentropy(from_logits=False)] * 4,
                      # # metrics=['categorical_accuracy'] * 4,
                      # metrics=[tf.keras.metrics.CategoricalAccuracy(name='categorical_accuracy'),
                      #          tf.keras.metrics.CategoricalAccuracy(name='categorical_accuracy_1'),
                      #          tf.keras.metrics.CategoricalAccuracy(name='categorical_accuracy_2'),
                      #          tf.keras.metrics.CategoricalAccuracy(name='categorical_accuracy_3')],
                      loss={'output_1': tf.keras.losses.CategoricalCrossentropy(from_logits=False),
                            'output_2': tf.keras.losses.CategoricalCrossentropy(from_logits=False),
                            'output_3': tf.keras.losses.CategoricalCrossentropy(from_logits=False),
                            'output_4': tf.keras.losses.CategoricalCrossentropy(from_logits=False)},
                      metrics={
                          'output_1': tf.keras.metrics.CategoricalAccuracy(name='acc'),
                          'output_2': tf.keras.metrics.CategoricalAccuracy(name='acc'),
                          'output_3': tf.keras.metrics.CategoricalAccuracy(name='acc'),
                          'output_4': tf.keras.metrics.CategoricalAccuracy(name='acc')},
                      loss_weights=[1.0, 1.0, 1.0, 1.0]
                      )