docs(device): warn against Kubernetes quantity suffixes for device memory requests

docs/userguide/cambricon-device/specify-device-memory-usage.md

@@ -13,6 +13,12 @@ resources:
     cambricon.com/mlu.smlu.vmemory: "20" # Each MLU contains 20% device memory
 ```
 
+:::warning
+
+Set `cambricon.com/mlu.smlu.vmemory` as a **plain integer** (here, a percentage from 1 to 100), not a Kubernetes quantity. A suffixed value such as `16Gi` is parsed as its byte count (about 17 billion), which overflows HAMi's 32-bit memory field and is silently truncated, often to `0`. A zero memory request passes the scheduler's memory check unconditionally, so the pod can be placed on an MLU that is already full and then fails with out-of-memory errors at runtime. Use the integer form shown above (for example `20`).
+
+:::
+
 :::note
 
 Depending on the parameters of cambricon-device-plugin, resource name can be `cambricon.com/mlu370.smlu.vmemory` or other types.

docs/userguide/hygon-device/specify-device-memory-usage.md

@@ -2,11 +2,17 @@
 title: Allocate device memory
 ---
 
-Allocate a percentage size of device memory by specifying resources such as `hygon.com/dcumem`. Optional, each unit of `hygon.com/dcumem` equals 1 MiB of device memory.
+Allocate device memory by specifying the `hygon.com/dcumem` resource. This field is optional. Each unit of `hygon.com/dcumem` represents 1 MiB of device memory.
 
 ```yaml
 resources:
   limits:
     hygon.com/dcunum: 1 # requesting 1 DCU
-    hygon.com/dcumem: 2000 # Each DCU contains 2000M device memory
+    hygon.com/dcumem: 2000 # 2000 units x 1 MiB = 2000 MiB of device memory
 ```
+
+:::warning
+
+Set `hygon.com/dcumem` as a **plain integer** (a count of MiB), not a Kubernetes quantity. A suffixed value such as `16Gi` is parsed as its byte count (about 17 billion), which overflows HAMi's 32-bit memory field and is silently truncated, often to `0`. A zero memory request passes the scheduler's memory check unconditionally, so the pod can be placed on a DCU that is already full and then fails with out-of-memory errors at runtime. Use the integer form shown above (for example `2000`).
+
+:::

docs/userguide/iluvatar-device/enable-iluvatar-gpu-sharing.md

@@ -121,6 +121,12 @@ Each unit of `iluvatar.ai/<card-type>.vMem` represents 256 MB of device memory.
 
 :::
 
+:::warning
+
+Set `iluvatar.ai/<card-type>.vMem` as a **plain integer** (a count of 256 MB units), not a Kubernetes quantity. A suffixed value such as `16Gi` is parsed as its byte count (about 17 billion), which overflows HAMi's 32-bit memory field and is silently truncated, often to `0`. A zero memory request passes the scheduler's memory check unconditionally, so the pod can be placed on a GPU that is already full and then fails with out-of-memory errors at runtime. Use the integer form shown above (for example `64`).
+
+:::
+
 ## Device UUID Selection
 
 You can specify which GPU devices to use or exclude using annotations:

docs/userguide/nvidia-device/specify-device-memory-usage.md

@@ -12,6 +12,12 @@ resources:
     nvidia.com/gpumem: 3000 # Each GPU contains 3000 MiB device memory
 ```
 
+:::warning
+
+Set `nvidia.com/gpumem` as a **plain integer** (a count of MiB), not a Kubernetes quantity. A suffixed value such as `16Gi` is parsed as its byte count (about 17 billion), which overflows HAMi's 32-bit memory field and is silently truncated, often to `0`. A zero memory request passes the scheduler's memory check unconditionally, so the pod can be placed on a GPU that is already full and then fails with out-of-memory errors at runtime. Use the integer form shown above (for example `3000`).
+
+:::
+
 Allocate a percentage of device memory by specifying resource `nvidia.com/gpumem-percentage`. Optional, each unit of `nvidia.com/gpumem-percentage` equals 1% of device memory.
 
 ```yaml

i18n/zh/docusaurus-plugin-content-docs/current/userguide/cambricon-device/specify-device-memory-usage.md

@@ -13,6 +13,12 @@ resources:
     cambricon.com/mlu.smlu.vmemory: "20" # 每个 MLU 包含 20% 的设备显存
 ```
 
+:::warning
+
+请将 `cambricon.com/mlu.smlu.vmemory` 设置为**纯整数**（此处为 1 到 100 的百分比），不要使用 Kubernetes 数量单位。带单位的值（例如 `16Gi`）会被解析为字节数（约 170 亿），从而超出 HAMi 的 32 位显存字段范围并被静默截断，通常截断为 `0`。显存请求为 0 时会无条件通过调度器的显存检查，导致 Pod 被调度到显存已满的设备上，并在运行时发生显存不足（OOM）。请使用上面示例中的整数形式（例如 `20`）。
+
+:::
+
 :::note
 
 根据 cambricon-device-plugin 的参数，资源名称可以是 `cambricon.com/mlu370.smlu.vmemory` 或其他类型

i18n/zh/docusaurus-plugin-content-docs/current/userguide/hygon-device/specify-device-memory-usage.md

@@ -4,11 +4,17 @@ sidebar_label: 指定显存
 translated: true
 ---
 
-通过指定诸如 `hygon.com/dcumem` 之类的资源来分配设备显存的百分比大小。可选项，每个 `hygon.com/dcumem` 单位等于 1M 设备显存。
+通过指定 `hygon.com/dcumem` 资源来分配设备显存。此字段为可选项，每个 `hygon.com/dcumem` 单位代表 1 MiB 设备显存。
 
 ```yaml
 resources:
   limits:
     hygon.com/dcunum: 1 # 请求 1 个 DCU
-    hygon.com/dcumem: 2000 # 每个 DCU 包含 2000M 设备显存
+    hygon.com/dcumem: 2000 # 2000 个单位 x 1 MiB = 2000 MiB 设备显存
 ```
+
+:::warning
+
+请将 `hygon.com/dcumem` 设置为**纯整数**（以 MiB 为单位计数），不要使用 Kubernetes 数量单位。带单位的值（例如 `16Gi`）会被解析为字节数（约 170 亿），从而超出 HAMi 的 32 位显存字段范围并被静默截断，通常截断为 `0`。显存请求为 0 时会无条件通过调度器的显存检查，导致 Pod 被调度到显存已满的设备上，并在运行时发生显存不足（OOM）。请使用上面示例中的整数形式（例如 `2000`）。
+
+:::

i18n/zh/docusaurus-plugin-content-docs/current/userguide/iluvatar-device/enable-iluvatar-gpu-sharing.md

@@ -117,6 +117,12 @@ spec:
 
 :::
 
+:::warning
+
+请将 `iluvatar.ai/<card-type>.vMem` 设置为**纯整数**（以 256MB 为单位计数），不要使用 Kubernetes 数量单位。带单位的值（例如 `16Gi`）会被解析为字节数（约 170 亿），从而超出 HAMi 的 32 位显存字段范围并被静默截断，通常截断为 `0`。显存请求为 0 时会无条件通过调度器的显存检查，导致 Pod 被调度到显存已满的设备上，并在运行时发生显存不足（OOM）。请使用上面示例中的整数形式（例如 `64`）。
+
+:::
+
 ## 设备 UUID 选择
 
 你可以通过 annotations 指定使用或排除哪些 GPU 设备：

i18n/zh/docusaurus-plugin-content-docs/current/userguide/nvidia-device/specify-device-memory-usage.md

@@ -13,6 +13,12 @@ resources:
     nvidia.com/gpumem: 3000 # 每个 GPU 包含 3000 MiB 设备显存
 ```
 
+:::warning
+
+请将 `nvidia.com/gpumem` 设置为**纯整数**（以 MiB 为单位计数），不要使用 Kubernetes 数量单位。带单位的值（例如 `16Gi`）会被解析为字节数（约 170 亿），从而超出 HAMi 的 32 位显存字段范围并被静默截断，通常截断为 `0`。显存请求为 0 时会无条件通过调度器的显存检查，导致 Pod 被调度到显存已满的设备上，并在运行时发生显存不足（OOM）。请使用上面示例中的整数形式（例如 `3000`）。
+
+:::
+
 通过指定资源 `nvidia.com/gpumem-percentage` 来分配设备显存的百分比。可选项，每个 `nvidia.com/gpumem-percentage` 单位等于设备显存的 1% 百分比。
 
 ```yaml

i18n/zh/docusaurus-plugin-content-docs/version-v2.9.0/userguide/cambricon-device/specify-device-memory-usage.md

@@ -13,6 +13,12 @@ resources:
     cambricon.com/mlu.smlu.vmemory: "20" # 每个 MLU 包含 20% 的设备显存
 ```
 
+:::warning
+
+请将 `cambricon.com/mlu.smlu.vmemory` 设置为**纯整数**（此处为 1 到 100 的百分比），不要使用 Kubernetes 数量单位。带单位的值（例如 `16Gi`）会被解析为字节数（约 170 亿），从而超出 HAMi 的 32 位显存字段范围并被静默截断，通常截断为 `0`。显存请求为 0 时会无条件通过调度器的显存检查，导致 Pod 被调度到显存已满的设备上，并在运行时发生显存不足（OOM）。请使用上面示例中的整数形式（例如 `20`）。
+
+:::
+
 :::note
 
 根据 cambricon-device-plugin 的参数，资源名称可以是 `cambricon.com/mlu370.smlu.vmemory` 或其他类型

i18n/zh/docusaurus-plugin-content-docs/version-v2.9.0/userguide/hygon-device/specify-device-memory-usage.md

@@ -4,11 +4,17 @@ sidebar_label: 指定显存
 translated: true
 ---
 
-通过指定诸如 `hygon.com/dcumem` 之类的资源来分配设备显存的百分比大小。可选项，每个 `hygon.com/dcumem` 单位等于 1M 设备显存。
+通过指定 `hygon.com/dcumem` 资源来分配设备显存。此字段为可选项，每个 `hygon.com/dcumem` 单位代表 1 MiB 设备显存。
 
 ```yaml
 resources:
   limits:
     hygon.com/dcunum: 1 # 请求 1 个 DCU
-    hygon.com/dcumem: 2000 # 每个 DCU 包含 2000M 设备显存
+    hygon.com/dcumem: 2000 # 2000 个单位 x 1 MiB = 2000 MiB 设备显存
 ```
+
+:::warning
+
+请将 `hygon.com/dcumem` 设置为**纯整数**（以 MiB 为单位计数），不要使用 Kubernetes 数量单位。带单位的值（例如 `16Gi`）会被解析为字节数（约 170 亿），从而超出 HAMi 的 32 位显存字段范围并被静默截断，通常截断为 `0`。显存请求为 0 时会无条件通过调度器的显存检查，导致 Pod 被调度到显存已满的设备上，并在运行时发生显存不足（OOM）。请使用上面示例中的整数形式（例如 `2000`）。
+
+:::

i18n/zh/docusaurus-plugin-content-docs/version-v2.9.0/userguide/iluvatar-device/enable-iluvatar-gpu-sharing.md

@@ -30,7 +30,7 @@ title: 启用天数智芯 GPU 共享
 - 安装 HAMi 时设置 `devices.iluvatar.enabled=true`
 
 ```bash
-helm install hami hami-charts/hami --set scheduler.kubeScheduler.imageTag={your kubernetes version} --set devices.iluvatar.enabled=true -n kube-system
+helm install hami hami-charts/hami --set scheduler.kubeScheduler.image.tag={your kubernetes version} --set devices.iluvatar.enabled=true -n kube-system
 ```
 
 **说明：** 当前支持的 GPU 型号及资源名称定义如下（位于 [https://github.com/Project-HAMi/HAMi/blob/master/charts/hami/templates/scheduler/device-configmap.yaml](https://github.com/Project-HAMi/HAMi/blob/master/charts/hami/templates/scheduler/device-configmap.yaml)）：
@@ -117,6 +117,12 @@ spec:
 
 :::
 
+:::warning
+
+请将 `iluvatar.ai/<card-type>.vMem` 设置为**纯整数**（以 256MB 为单位计数），不要使用 Kubernetes 数量单位。带单位的值（例如 `16Gi`）会被解析为字节数（约 170 亿），从而超出 HAMi 的 32 位显存字段范围并被静默截断，通常截断为 `0`。显存请求为 0 时会无条件通过调度器的显存检查，导致 Pod 被调度到显存已满的设备上，并在运行时发生显存不足（OOM）。请使用上面示例中的整数形式（例如 `64`）。
+
+:::
+
 ## 设备 UUID 选择
 
 你可以通过 annotations 指定使用或排除哪些 GPU 设备：

i18n/zh/docusaurus-plugin-content-docs/version-v2.9.0/userguide/nvidia-device/specify-device-memory-usage.md

@@ -10,9 +10,15 @@ translated: true
 resources:
   limits:
     nvidia.com/gpu: 1 # 请求 1 个 GPU
-    nvidia.com/gpumem: 3000 # 每个 GPU 包含 3000m 设备显存
+    nvidia.com/gpumem: 3000 # 每个 GPU 包含 3000 MiB 设备显存
 ```
 
+:::warning
+
+请将 `nvidia.com/gpumem` 设置为**纯整数**（以 MiB 为单位计数），不要使用 Kubernetes 数量单位。带单位的值（例如 `16Gi`）会被解析为字节数（约 170 亿），从而超出 HAMi 的 32 位显存字段范围并被静默截断，通常截断为 `0`。显存请求为 0 时会无条件通过调度器的显存检查，导致 Pod 被调度到显存已满的设备上，并在运行时发生显存不足（OOM）。请使用上面示例中的整数形式（例如 `3000`）。
+
+:::
+
 通过指定资源 `nvidia.com/gpumem-percentage` 来分配设备显存的百分比。可选项，每个 `nvidia.com/gpumem-percentage` 单位等于设备显存的 1% 百分比。
 
 ```yaml

versioned_docs/version-v2.9.0/userguide/cambricon-device/specify-device-memory-usage.md

@@ -13,6 +13,12 @@ resources:
     cambricon.com/mlu.smlu.vmemory: "20" # Each MLU contains 20% device memory
 ```
 
+:::warning
+
+Set `cambricon.com/mlu.smlu.vmemory` as a **plain integer** (here, a percentage from 1 to 100), not a Kubernetes quantity. A suffixed value such as `16Gi` is parsed as its byte count (about 17 billion), which overflows HAMi's 32-bit memory field and is silently truncated, often to `0`. A zero memory request passes the scheduler's memory check unconditionally, so the pod can be placed on an MLU that is already full and then fails with out-of-memory errors at runtime. Use the integer form shown above (for example `20`).
+
+:::
+
 :::note
 
 Depending on the parameters of cambricon-device-plugin, resource name can be `cambricon.com/mlu370.smlu.vmemory` or other types.

versioned_docs/version-v2.9.0/userguide/hygon-device/specify-device-memory-usage.md

@@ -2,11 +2,17 @@
 title: Allocate device memory
 ---
 
-Allocate a percentage size of device memory by specifying resources such as `hygon.com/dcumem`. Optional, each unit of `hygon.com/dcumem` equals 1 MiB of device memory.
+Allocate device memory by specifying the `hygon.com/dcumem` resource. This field is optional. Each unit of `hygon.com/dcumem` represents 1 MiB of device memory.
 
 ```yaml
 resources:
   limits:
     hygon.com/dcunum: 1 # requesting 1 DCU
-    hygon.com/dcumem: 2000 # Each DCU contains 2000M device memory
+    hygon.com/dcumem: 2000 # 2000 units x 1 MiB = 2000 MiB of device memory
 ```
+
+:::warning
+
+Set `hygon.com/dcumem` as a **plain integer** (a count of MiB), not a Kubernetes quantity. A suffixed value such as `16Gi` is parsed as its byte count (about 17 billion), which overflows HAMi's 32-bit memory field and is silently truncated, often to `0`. A zero memory request passes the scheduler's memory check unconditionally, so the pod can be placed on a DCU that is already full and then fails with out-of-memory errors at runtime. Use the integer form shown above (for example `2000`).
+
+:::

versioned_docs/version-v2.9.0/userguide/iluvatar-device/enable-iluvatar-gpu-sharing.md

@@ -34,7 +34,7 @@ Install only gpu-manager. Do not install the gpu-admission package.
 - Set `devices.iluvatar.enabled=true` when installing HAMi
 
 ```bash
-helm install hami hami-charts/hami --set scheduler.kubeScheduler.imageTag={your kubernetes version} --set devices.iluvatar.enabled=true -n kube-system
+helm install hami hami-charts/hami --set scheduler.kubeScheduler.image.tag={your kubernetes version} --set devices.iluvatar.enabled=true -n kube-system
 ```
 
 The currently supported GPU models and resource names are defined in [device-configmap.yaml](https://github.com/Project-HAMi/HAMi/blob/master/charts/hami/templates/scheduler/device-configmap.yaml):
@@ -121,6 +121,12 @@ Each unit of `iluvatar.ai/<card-type>.vMem` represents 256 MB of device memory.
 
 :::
 
+:::warning
+
+Set `iluvatar.ai/<card-type>.vMem` as a **plain integer** (a count of 256 MB units), not a Kubernetes quantity. A suffixed value such as `16Gi` is parsed as its byte count (about 17 billion), which overflows HAMi's 32-bit memory field and is silently truncated, often to `0`. A zero memory request passes the scheduler's memory check unconditionally, so the pod can be placed on a GPU that is already full and then fails with out-of-memory errors at runtime. Use the integer form shown above (for example `64`).
+
+:::
+
 ## Device UUID Selection
 
 You can specify which GPU devices to use or exclude using annotations:

versioned_docs/version-v2.9.0/userguide/nvidia-device/specify-device-memory-usage.md

@@ -9,9 +9,15 @@ Allocate a certain size of device memory by specifying resources such as `nvidia
 resources:
   limits:
     nvidia.com/gpu: 1 # requesting 1 GPU
-    nvidia.com/gpumem: 3000 # Each GPU contains 3000m device memory
+    nvidia.com/gpumem: 3000 # Each GPU contains 3000 MiB device memory
 ```
 
+:::warning
+
+Set `nvidia.com/gpumem` as a **plain integer** (a count of MiB), not a Kubernetes quantity. A suffixed value such as `16Gi` is parsed as its byte count (about 17 billion), which overflows HAMi's 32-bit memory field and is silently truncated, often to `0`. A zero memory request passes the scheduler's memory check unconditionally, so the pod can be placed on a GPU that is already full and then fails with out-of-memory errors at runtime. Use the integer form shown above (for example `3000`).
+
+:::
+
 Allocate a percentage of device memory by specifying resource `nvidia.com/gpumem-percentage`. Optional, each unit of `nvidia.com/gpumem-percentage` equals 1% of device memory.
 
 ```yaml
@@ -23,6 +29,6 @@ resources:
 
 :::note
 
-`nvidia.com/gpumem` and `nvidia.com/gpumem-percentage` can't be assigned together
+`nvidia.com/gpumem` and `nvidia.com/gpumem-percentage` cannot be assigned together
 
 :::