],
)

tf_cuda_cc_test(
    name = "gradients_test",
    size = "small",
    srcs = [
        "gradients_test.cc",
    ],
    args = ["--heap_check="],
    tags = tf_cuda_tests_tags() + ["nomac"],
    deps = [
        ":abstract_context",
        ":abstract_tensor_handle",
        ":c_api_experimental",
        ":c_api_test_util",
        ":c_api_unified_internal",
        ":gradients_internal",

test:macos_arm64_wheel_test_filters --test_tag_filters=-no_oss,-tf_tosa,-oss_excluded,-oss_serial,-no_oss_py310,-no_oss_py313,-nomac,-no_mac,-mac_excluded,-v1only,-gpu,-tpu,-benchmark-test,-no_mac_arm64,-no_aarch64
test:macos_arm64_wheel_test_filters --build_tag_filters=-no_oss,-tf_tosa,-oss_excluded,-oss_serial,-no_oss_py310,-no_oss_py313,-nomac,-no_mac,-mac_excluded,-v1only,-gpu,-tpu,-benchmark-test,-no_mac_arm64,-no_aarch64

      test_and_build_filters += ['-no_windows_gpu', '-no_gpu']
    else:
      test_and_build_filters.append('-gpu')
  elif is_macos():
    test_and_build_filters += ['-gpu', '-nomac', '-no_mac', '-mac_excluded']
  elif is_linux():
    if ((environ_cp.get('TF_NEED_CUDA', None) == '1') or
        (environ_cp.get('TF_NEED_ROCM', None) == '1')):
      test_and_build_filters.append('-no_gpu')

processus { #replication-number-of-processes }

Si vous avez un <dfn title="Groupe de machines configurées pour être connectées et fonctionner ensemble d'une certaine manière.">cluster</dfn> de machines avec **Kubernetes**, Docker Swarm Mode, Nomad, ou un autre système complexe similaire pour gérer des conteneurs distribués sur plusieurs machines, alors vous voudrez probablement **gérer la réplication** au **niveau du cluster** plutôt que d'utiliser un **gestionnaire de processus** (comme Uvicorn...

количество процессов { #replication-number-of-processes }

Если у вас есть <dfn title="Группа машин, настроенных так, чтобы быть соединенными и работать вместе определенным образом.">кластер</dfn> машин с **Kubernetes**, Docker Swarm Mode, Nomad или другой похожей системой для управления распределёнными контейнерами на нескольких машинах, скорее всего вы будете **управлять репликацией** на **уровне кластера**, а не использовать **менеджер процессов** (например, Uvicorn с воркерами) в каждом...

}

Eğer bir <dfn title="Bir şekilde birbirine bağlanacak ve birlikte çalışacak şekilde yapılandırılmış makineler grubu.">küme</dfn> (cluster) olarak yapılandırılmış makineler grubunuz varsa ve bunları **Kubernetes**, Docker Swarm Mode, Nomad veya benzeri, birden çok makinede dağıtık container'ları yöneten karmaşık bir sistemle yönetiyorsanız, replication'ı her container içinde bir **process manager** (ör. worker'lı Uvicorn) kullanarak yönetmek yerine, muhtemelen **küme seviyesinde (cluster...

Number of Processes { #replication-number-of-processes }

If you have a <dfn title="A group of machines that are configured to be connected and work together in some way.">cluster</dfn> of machines with **Kubernetes**, Docker Swarm Mode, Nomad, or another similar complex system to manage distributed containers on multiple machines, then you will probably want to **handle replication** at the **cluster level** instead of using a **process manager** (like Uvicorn with workers) in each container....

## Реплікація - кількість процесів { #replication-number-of-processes }

{ #replication-number-of-processes }

Se você tiver um <dfn title="Um grupo de máquinas que são configuradas para estarem conectadas e trabalharem juntas de alguma forma.">cluster</dfn> de máquinas com **Kubernetes**, Docker Swarm Mode, Nomad ou outro sistema complexo semelhante para gerenciar contêineres distribuídos em várias máquinas, então provavelmente desejará **lidar com a replicação** no **nível do cluster** em vez de usar um **gerenciador de processos** (como Uvicorn com workers)...

在大多数（或全部）情况下，都有简单选项可以在开机时运行容器并在失败时启用重启。例如，在 Docker 中是命令行选项 `--restart`。

如果不使用容器，要让应用开机自启并带重启可能繁琐且困难。但在**容器**场景下，这种功能通常默认就包含了。✨

## 复制 - 进程数 { #replication-number-of-processes }

如果你有一个由 **Kubernetes**、Docker Swarm Mode、Nomad 或其他类似的复杂系统管理的、在多台机器上运行的分布式容器<dfn title="被配置为以某种方式连接并协同工作的多台机器">集群</dfn>，那么你很可能会希望在**集群层面**来**处理复制**，而不是在每个容器中使用**进程管理**（比如让 Uvicorn 运行多个 workers）。

像 Kubernetes 这样的分布式容器管理系统通常都有某种内置方式来处理**容器复制**，同时对传入请求进行**负载均衡**。这一切都在**集群层面**完成。