diff --git a/docs/en/kubeflow/how_to/assets/training-runtimes/llamafactory0.9-cann8.5-arm64-trainingruntime.yaml b/docs/en/kubeflow/how_to/assets/training-runtimes/llamafactory0.9-cann8.5-arm64-trainingruntime.yaml
new file mode 100644
index 0000000..e0b0e02
--- /dev/null
+++ b/docs/en/kubeflow/how_to/assets/training-runtimes/llamafactory0.9-cann8.5-arm64-trainingruntime.yaml
@@ -0,0 +1,67 @@
+apiVersion: trainer.kubeflow.org/v1alpha1
+kind: TrainingRuntime
+metadata:
+  name: llamafactory0.9-cann8.5-arm64
+  namespace: kubeflow-admin-cpaas-io
+  labels:
+    trainer.kubeflow.org/framework: torch
+    alauda.io/training-runtime: llamafactory0.9-cann8.5-arm64
+spec:
+  mlPolicy:
+    numNodes: 1
+    torch:
+      numProcPerNode: auto
+  template:
+    spec:
+      replicatedJobs:
+        - name: node
+          template:
+            metadata:
+              labels:
+                trainer.kubeflow.org/trainjob-ancestor-step: trainer
+            spec:
+              backoffLimit: 0
+              template:
+                spec:
+                  schedulerName: hami-scheduler
+                  runtimeClassName: ascend
+                  securityContext:
+                    runAsNonRoot: true
+                    runAsUser: 1001
+                    runAsGroup: 1001
+                    fsGroup: 1001
+                  containers:
+                    - name: node
+                      image: docker.io/alaudadockerhub/llamafactory0.9-cann8.5-arm64:v0.1.0
+                      command:
+                        - bash
+                        - -c
+                        - |
+                          set -ex
+                          python -c "import torch, torch_npu, llamafactory; print('torch', torch.__version__, 'torch_npu', torch_npu.__version__, 'lf', llamafactory.__version__, 'npu_available', torch_npu.npu.is_available())"
+                          llamafactory-cli version || true
+                      resources:
+                        requests:
+                          cpu: "1"
+                          memory: 4Gi
+                        limits:
+                          cpu: "4"
+                          memory: 8Gi
+                          huawei.com/Ascend910B4: "1"
+                          huawei.com/Ascend910B4-memory: "8192"
+                      securityContext:
+                        allowPrivilegeEscalation: false
+                        capabilities:
+                          drop: [ALL]
+                        runAsNonRoot: true
+                        seccompProfile:
+                          type: RuntimeDefault
+                      volumeMounts:
+                        - { name: workspace, mountPath: /workspace }
+                        - { name: dshm, mountPath: /dev/shm }
+                  volumes:
+                    - { name: workspace, emptyDir: {} }
+                    - name: dshm
+                      emptyDir:
+                        medium: Memory
+                        sizeLimit: 2Gi
diff --git a/docs/en/kubeflow/how_to/assets/training-runtimes/llamafactory0.9-cu126-amd64-trainingruntime.yaml b/docs/en/kubeflow/how_to/assets/training-runtimes/llamafactory0.9-cu126-amd64-trainingruntime.yaml
new file mode 100644
index 0000000..aeb765e
--- /dev/null
+++ b/docs/en/kubeflow/how_to/assets/training-runtimes/llamafactory0.9-cu126-amd64-trainingruntime.yaml
@@ -0,0 +1,81 @@
+apiVersion: trainer.kubeflow.org/v1alpha1
+kind: TrainingRuntime
+metadata:
+  name: llamafactory0.9-cu126-amd64
+  namespace: kubeflow-admin-cpaas-io
+  labels:
+    trainer.kubeflow.org/framework: torch
+    alauda.io/training-runtime: llamafactory0.9-cu126-amd64
+spec:
+  mlPolicy:
+    numNodes: 1
+    torch:
+      numProcPerNode: auto
+  template:
+    spec:
+      replicatedJobs:
+        - name: node
+          template:
+            metadata:
+              labels:
+                trainer.kubeflow.org/trainjob-ancestor-step: trainer
+            spec:
+              backoffLimit: 0
+              template:
+                spec:
+                  securityContext:
+                    runAsNonRoot: true
+                    runAsUser: 1000
+                    runAsGroup: 1000
+                    fsGroup: 1000
+                  containers:
+                    - name: node
+                      image: docker.io/alaudadockerhub/llamafactory0.9-cu126-amd64:v0.1.0
+                      command:
+                        - bash
+                        - -c
+                        - |
+                          set -ex
+                          # Synthetic identity SFT smoke. Writes a 4-row alpaca dataset, fine-tunes
+                          # qwen-style chat template against an empty config to exercise the
+                          # LLaMA-Factory CLI without downloading a real model.
+                          mkdir -p /tmp/lf
+                          cat >/tmp/lf/dataset_info.json <<'JSON'
+                          {"identity_smoke": {"file_name": "identity_smoke.json"}}
+                          JSON
+                          cat >/tmp/lf/identity_smoke.json <<'JSON'
+                          [
+                            {"instruction":"Who are you?","input":"","output":"I am Alauda AI."},
+                            {"instruction":"Where do you run?","input":"","output":"On Kubeflow Trainer v2."},
+                            {"instruction":"What framework powers you?","input":"","output":"LLaMA-Factory on PyTorch."},
+                            {"instruction":"What device?","input":"","output":"NVIDIA GPU via CUDA 12.6."}
+                          ]
+                          JSON
+                          python -c "import llamafactory, torch; print('lf', llamafactory.__version__, 'torch', torch.__version__, 'cuda', torch.cuda.is_available())"
+                          llamafactory-cli version || true
+                      resources:
+                        requests:
+                          cpu: "1"
+                          memory: 4Gi
+                        limits:
+                          cpu: "4"
+                          memory: 8Gi
+                          nvidia.com/gpualloc: 1
+                          nvidia.com/gpucores: 50
+                          nvidia.com/gpumem: "8192"
+                      securityContext:
+                        allowPrivilegeEscalation: false
+                        capabilities:
+                          drop: [ALL]
+                        runAsNonRoot: true
+                        seccompProfile:
+                          type: RuntimeDefault
+                      volumeMounts:
+                        - { name: workspace, mountPath: /workspace }
+                        - { name: dshm, mountPath: /dev/shm }
+                  volumes:
+                    - { name: workspace, emptyDir: {} }
+                    - name: dshm
+                      emptyDir:
+                        medium: Memory
+                        sizeLimit: 2Gi
diff --git a/docs/en/kubeflow/how_to/assets/training-runtimes/mindspeed-llm-cann8.5-arm64-trainingruntime.yaml b/docs/en/kubeflow/how_to/assets/training-runtimes/mindspeed-llm-cann8.5-arm64-trainingruntime.yaml
new file mode 100644
index 0000000..b212e3c
--- /dev/null
+++ b/docs/en/kubeflow/how_to/assets/training-runtimes/mindspeed-llm-cann8.5-arm64-trainingruntime.yaml
@@ -0,0 +1,66 @@
+apiVersion: trainer.kubeflow.org/v1alpha1
+kind: TrainingRuntime
+metadata:
+  name: mindspeed-llm-cann8.5-arm64
+  namespace: kubeflow-admin-cpaas-io
+  labels:
+    trainer.kubeflow.org/framework: torch
+    alauda.io/training-runtime: mindspeed-llm-cann8.5-arm64
+spec:
+  mlPolicy:
+    numNodes: 1
+    torch:
+      numProcPerNode: auto
+  template:
+    spec:
+      replicatedJobs:
+        - name: node
+          template:
+            metadata:
+              labels:
+                trainer.kubeflow.org/trainjob-ancestor-step: trainer
+            spec:
+              backoffLimit: 0
+              template:
+                spec:
+                  schedulerName: hami-scheduler
+                  runtimeClassName: ascend
+                  securityContext:
+                    runAsNonRoot: true
+                    runAsUser: 1001
+                    runAsGroup: 1001
+                    fsGroup: 1001
+                  containers:
+                    - name: node
+                      image: docker.io/alaudadockerhub/mindspeed-llm-cann8.5-arm64:v0.1.0
+                      command:
+                        - bash
+                        - -c
+                        - |
+                          set -ex
+                          python -c "import torch, torch_npu, megatron, mindspeed, mindspeed_llm; print('torch', torch.__version__, 'torch_npu', torch_npu.__version__, 'megatron', megatron.__version__, 'mindspeed_llm', mindspeed_llm.__version__)"
+                      resources:
+                        requests:
+                          cpu: "1"
+                          memory: 4Gi
+                        limits:
+                          cpu: "4"
+                          memory: 8Gi
+                          huawei.com/Ascend910B4: "1"
+                          huawei.com/Ascend910B4-memory: "8192"
+                      securityContext:
+                        allowPrivilegeEscalation: false
+                        capabilities:
+                          drop: [ALL]
+                        runAsNonRoot: true
+                        seccompProfile:
+                          type: RuntimeDefault
+                      volumeMounts:
+                        - { name: workspace, mountPath: /workspace }
+                        - { name: dshm, mountPath: /dev/shm }
+                  volumes:
+                    - { name: workspace, emptyDir: {} }
+                    - name: dshm
+                      emptyDir:
+                        medium: Memory
+                        sizeLimit: 2Gi
diff --git a/docs/en/kubeflow/how_to/assets/training-runtimes/torch2.6-cann8.5-arm64-trainingruntime.yaml b/docs/en/kubeflow/how_to/assets/training-runtimes/torch2.6-cann8.5-arm64-trainingruntime.yaml
new file mode 100644
index 0000000..493cc90
--- /dev/null
+++ b/docs/en/kubeflow/how_to/assets/training-runtimes/torch2.6-cann8.5-arm64-trainingruntime.yaml
@@ -0,0 +1,77 @@
+apiVersion: trainer.kubeflow.org/v1alpha1
+kind: TrainingRuntime
+metadata:
+  name: torch2.6-cann8.5-arm64
+  namespace: kubeflow-admin-cpaas-io
+  labels:
+    trainer.kubeflow.org/framework: torch
+    alauda.io/training-runtime: torch2.6-cann8.5-arm64
+spec:
+  mlPolicy:
+    numNodes: 1
+    torch:
+      numProcPerNode: auto
+  template:
+    spec:
+      replicatedJobs:
+        - name: node
+          template:
+            metadata:
+              labels:
+                trainer.kubeflow.org/trainjob-ancestor-step: trainer
+            spec:
+              backoffLimit: 0
+              template:
+                spec:
+                  schedulerName: hami-scheduler
+                  runtimeClassName: ascend
+                  securityContext:
+                    runAsNonRoot: true
+                    runAsUser: 1001
+                    runAsGroup: 1001
+                    fsGroup: 1001
+                  containers:
+                    - name: node
+                      image: docker.io/alaudadockerhub/torch2.6-cann8.5-arm64:v0.1.0
+                      command:
+                        - python
+                        - -c
+                        - |
+                          import os, torch
+                          try:
+                              import torch_npu
+                              import torch_npu.utils
+                              print(f"torch={torch.__version__} torch_npu={torch_npu.__version__}")
+                              print(f"npu_available={torch_npu.npu.is_available()} count={torch_npu.npu.device_count()}")
+                              if torch_npu.npu.is_available():
+                                  x = torch.randn(1024, 1024).npu()
+                                  y = x @ x.T
+                                  print(f"matmul ok shape={tuple(y.shape)} mean={y.mean().item():.4f}")
+                          except Exception as e:
+                              print(f"torch_npu probe failed: {e}")
+                              raise
+                      resources:
+                        requests:
+                          cpu: "1"
+                          memory: 2Gi
+                        limits:
+                          cpu: "4"
+                          memory: 8Gi
+                          huawei.com/Ascend910B4: "1"
+                          huawei.com/Ascend910B4-memory: "8192"
+                      securityContext:
+                        allowPrivilegeEscalation: false
+                        capabilities:
+                          drop: [ALL]
+                        runAsNonRoot: true
+                        seccompProfile:
+                          type: RuntimeDefault
+                      volumeMounts:
+                        - { name: workspace, mountPath: /workspace }
+                        - { name: dshm, mountPath: /dev/shm }
+                  volumes:
+                    - { name: workspace, emptyDir: {} }
+                    - name: dshm
+                      emptyDir:
+                        medium: Memory
+                        sizeLimit: 1Gi
diff --git a/docs/en/kubeflow/how_to/assets/training-runtimes/torch2.6-cu126-amd64-trainingruntime.yaml b/docs/en/kubeflow/how_to/assets/training-runtimes/torch2.6-cu126-amd64-trainingruntime.yaml
new file mode 100644
index 0000000..aabae60
--- /dev/null
+++ b/docs/en/kubeflow/how_to/assets/training-runtimes/torch2.6-cu126-amd64-trainingruntime.yaml
@@ -0,0 +1,70 @@
+apiVersion: trainer.kubeflow.org/v1alpha1
+kind: TrainingRuntime
+metadata:
+  name: torch2.6-cu126-amd64
+  namespace: kubeflow-admin-cpaas-io
+  labels:
+    trainer.kubeflow.org/framework: torch
+    alauda.io/training-runtime: torch2.6-cu126-amd64
+spec:
+  mlPolicy:
+    numNodes: 1
+    torch:
+      numProcPerNode: auto
+  template:
+    spec:
+      replicatedJobs:
+        - name: node
+          template:
+            metadata:
+              labels:
+                trainer.kubeflow.org/trainjob-ancestor-step: trainer
+            spec:
+              backoffLimit: 0
+              template:
+                spec:
+                  securityContext:
+                    runAsNonRoot: true
+                    runAsUser: 1000
+                    runAsGroup: 1000
+                    fsGroup: 1000
+                  containers:
+                    - name: node
+                      image: docker.io/alaudadockerhub/torch2.6-cu126-amd64:v0.1.0
+                      command:
+                        - python
+                        - -c
+                        - |
+                          import torch, os
+                          print(f"torch={torch.__version__} cuda={torch.cuda.is_available()} devices={torch.cuda.device_count()}")
+                          print(f"PET_NNODES={os.environ.get('PET_NNODES')} PET_NODE_RANK={os.environ.get('PET_NODE_RANK')}")
+                          if torch.cuda.is_available():
+                              x = torch.randn(1024, 1024, device='cuda')
+                              y = x @ x.T
+                              print(f"matmul ok, shape={tuple(y.shape)}, mean={y.mean().item():.4f}")
+                      resources:
+                        requests:
+                          cpu: "1"
+                          memory: 2Gi
+                        limits:
+                          cpu: "2"
+                          memory: 4Gi
+                          nvidia.com/gpualloc: 1
+                          nvidia.com/gpucores: 50
+                          nvidia.com/gpumem: "8192"
+                      securityContext:
+                        allowPrivilegeEscalation: false
+                        capabilities:
+                          drop: [ALL]
+                        runAsNonRoot: true
+                        seccompProfile:
+                          type: RuntimeDefault
+                      volumeMounts:
+                        - { name: workspace, mountPath: /workspace }
+                        - { name: dshm, mountPath: /dev/shm }
+                  volumes:
+                    - { name: workspace, emptyDir: {} }
+                    - name: dshm
+                      emptyDir:
+                        medium: Memory
+                        sizeLimit: 1Gi
diff --git a/docs/en/kubeflow/how_to/assets/training-runtimes/traininghub0.1-cu126-amd64-trainingruntime.yaml b/docs/en/kubeflow/how_to/assets/training-runtimes/traininghub0.1-cu126-amd64-trainingruntime.yaml
new file mode 100644
index 0000000..72a3064
--- /dev/null
+++ b/docs/en/kubeflow/how_to/assets/training-runtimes/traininghub0.1-cu126-amd64-trainingruntime.yaml
@@ -0,0 +1,70 @@
+apiVersion: trainer.kubeflow.org/v1alpha1
+kind: TrainingRuntime
+metadata:
+  name: traininghub0.1-cu126-amd64
+  namespace: kubeflow-admin-cpaas-io
+  labels:
+    trainer.kubeflow.org/framework: torch
+    alauda.io/training-runtime: traininghub0.1-cu126-amd64
+spec:
+  mlPolicy:
+    numNodes: 1
+    torch:
+      numProcPerNode: auto
+  template:
+    spec:
+      replicatedJobs:
+        - name: node
+          template:
+            metadata:
+              labels:
+                trainer.kubeflow.org/trainjob-ancestor-step: trainer
+            spec:
+              backoffLimit: 0
+              template:
+                spec:
+                  securityContext:
+                    runAsNonRoot: true
+                    runAsUser: 1000
+                    runAsGroup: 1000
+                    fsGroup: 1000
+                  containers:
+                    - name: node
+                      image: docker.io/alaudadockerhub/traininghub0.1-cu126-amd64:v0.1.0
+                      command:
+                        - python
+                        - -c
+                        - |
+                          import torch, trl, peft, bitsandbytes, deepspeed
+                          print(f"torch={torch.__version__} cuda={torch.cuda.is_available()}")
+                          print(f"trl={trl.__version__} peft={peft.__version__} bnb={bitsandbytes.__version__} deepspeed={deepspeed.__version__}")
+                          if torch.cuda.is_available():
+                              from peft import LoraConfig
+                              cfg = LoraConfig(r=4, lora_alpha=8, lora_dropout=0.05, target_modules=["q_proj","v_proj"])
+                              print("peft LoraConfig ok", cfg)
+                      resources:
+                        requests:
+                          cpu: "1"
+                          memory: 4Gi
+                        limits:
+                          cpu: "4"
+                          memory: 8Gi
+                          nvidia.com/gpualloc: 1
+                          nvidia.com/gpucores: 50
+                          nvidia.com/gpumem: "8192"
+                      securityContext:
+                        allowPrivilegeEscalation: false
+                        capabilities:
+                          drop: [ALL]
+                        runAsNonRoot: true
+                        seccompProfile:
+                          type: RuntimeDefault
+                      volumeMounts:
+                        - { name: workspace, mountPath: /workspace }
+                        - { name: dshm, mountPath: /dev/shm }
+                  volumes:
+                    - { name: workspace, emptyDir: {} }
+                    - name: dshm
+                      emptyDir:
+                        medium: Memory
+                        sizeLimit: 2Gi
diff --git a/docs/en/kubeflow/how_to/assets/training-runtimes/trainjob-smoke.yaml b/docs/en/kubeflow/how_to/assets/training-runtimes/trainjob-smoke.yaml
new file mode 100644
index 0000000..4fcd6a0
--- /dev/null
+++ b/docs/en/kubeflow/how_to/assets/training-runtimes/trainjob-smoke.yaml
@@ -0,0 +1,21 @@
+# Smoke TrainJob that exercises any of the published TrainingRuntimes.
+# Set `spec.runtimeRef.name` to one of:
+#   torch2.6-cu126-amd64
+#   torch2.6-cann8.5-arm64
+#   llamafactory0.9-cu126-amd64
+#   llamafactory0.9-cann8.5-arm64
+#   traininghub0.1-cu126-amd64
+#   mindspeed-llm-cann8.5-arm64
+apiVersion: trainer.kubeflow.org/v1alpha1
+kind: TrainJob
+metadata:
+  generateName: smoke-
+  namespace: kubeflow-admin-cpaas-io
+spec:
+  runtimeRef:
+    apiGroup: trainer.kubeflow.org
+    kind: TrainingRuntime
+    name: torch2.6-cu126-amd64
+  suspend: false
+  trainer:
+    numNodes: 1
diff --git a/docs/en/kubeflow/how_to/training-runtimes.mdx b/docs/en/kubeflow/how_to/training-runtimes.mdx
new file mode 100644
index 0000000..8631df6
--- /dev/null
+++ b/docs/en/kubeflow/how_to/training-runtimes.mdx
@@ -0,0 +1,175 @@
+---
+weight: 85
+---
+
+# Training Runtime Images
+
+Alauda AI ships a curated catalog of [Kubeflow Trainer v2](./fine-tune-with-trainer-v2.mdx)
+`TrainingRuntime` images. Each image bundles a specific PyTorch + accelerator
+stack (NVIDIA CUDA or Huawei Ascend NPU via CANN) so users can submit
+`TrainJob`s without rebuilding a custom image.
+
+## Available runtimes
+
+The published `v0.1.0` images on Dockerhub:
+
+| Image | Device | Framework | Pull |
+|---|---|---|---|
+| torch (CUDA) | NVIDIA GPU (CUDA 12.6) | PyTorch 2.6, transformers, accelerate, datasets, mlflow | `alaudadockerhub/torch2.6-cu126-amd64:v0.1.0` |
+| torch (CANN) | Huawei Ascend NPU (CANN 8.5) | PyTorch 2.6 + torch_npu 2.6.0.post5 | `alaudadockerhub/torch2.6-cann8.5-arm64:v0.1.0` |
+| LLaMA-Factory (CUDA) | NVIDIA GPU (CUDA 12.6) | LLaMA-Factory 0.9.4 (`metrics,awq,modelscope`) | `alaudadockerhub/llamafactory0.9-cu126-amd64:v0.1.0` |
+| LLaMA-Factory (CANN) | Huawei Ascend NPU (CANN 8.5) | LLaMA-Factory 0.9.4 (`metrics,modelscope`, no `awq`) | `alaudadockerhub/llamafactory0.9-cann8.5-arm64:v0.1.0` |
+| TrainingHub (CUDA) | NVIDIA GPU (CUDA 12.6) | trl, peft, bitsandbytes, deepspeed (SFT/OSFT/DPO) | `alaudadockerhub/traininghub0.1-cu126-amd64:v0.1.0` |
+| MindSpeed-LLM (CANN) | Huawei Ascend NPU (CANN 8.5) | MindSpeed + MindSpeed-LLM (Megatron core 0.8.0) | `alaudadockerhub/mindspeed-llm-cann8.5-arm64:v0.1.0` |
+
+CUDA images are amd64-only — NVIDIA does not ship CUDA wheels for arm64.
+CANN images are arm64-only — the upstream Ascend CANN base image is
+published for arm64 only.
+
+Pull example:
+
+```bash
+docker pull alaudadockerhub/torch2.6-cu126-amd64:v0.1.0
+```
+
+## Picking a runtime
+
+- **Plain torchrun training** on NVIDIA GPUs → `torch2.6-cu126-amd64`.
+- **Plain torchrun training** on Huawei Ascend NPUs → `torch2.6-cann8.5-arm64`
+  (set `runtimeClassName: ascend` on the pod template).
+- **LLM SFT / LoRA fine-tuning with LLaMA-Factory** →
+  `llamafactory0.9-cu126-amd64` on GPU or `llamafactory0.9-cann8.5-arm64`
+  on NPU. See [Fine-Tuning with Kubeflow Trainer v2](./fine-tune-with-trainer-v2.mdx)
+  for the dataset / model initializer pipeline.
+- **TRL / PEFT-based SFT / OSFT / DPO** → `traininghub0.1-cu126-amd64`.
+- **Megatron-style training on Ascend** → `mindspeed-llm-cann8.5-arm64`.
+
+## TrainingRuntime YAML examples
+
+Ready-to-apply YAMLs for each runtime are shipped under
+[`assets/training-runtimes/`](https://github.com/alauda/aml-docs/tree/master/docs/en/kubeflow/how_to/assets/training-runtimes):
+
+```bash
+# NVIDIA GPU
+kubectl apply -f https://raw.githubusercontent.com/alauda/aml-docs/master/docs/en/kubeflow/how_to/assets/training-runtimes/torch2.6-cu126-amd64-trainingruntime.yaml
+kubectl apply -f https://raw.githubusercontent.com/alauda/aml-docs/master/docs/en/kubeflow/how_to/assets/training-runtimes/llamafactory0.9-cu126-amd64-trainingruntime.yaml
+kubectl apply -f https://raw.githubusercontent.com/alauda/aml-docs/master/docs/en/kubeflow/how_to/assets/training-runtimes/traininghub0.1-cu126-amd64-trainingruntime.yaml
+
+# Huawei Ascend NPU
+kubectl apply -f https://raw.githubusercontent.com/alauda/aml-docs/master/docs/en/kubeflow/how_to/assets/training-runtimes/torch2.6-cann8.5-arm64-trainingruntime.yaml
+kubectl apply -f https://raw.githubusercontent.com/alauda/aml-docs/master/docs/en/kubeflow/how_to/assets/training-runtimes/llamafactory0.9-cann8.5-arm64-trainingruntime.yaml
+kubectl apply -f https://raw.githubusercontent.com/alauda/aml-docs/master/docs/en/kubeflow/how_to/assets/training-runtimes/mindspeed-llm-cann8.5-arm64-trainingruntime.yaml
+```
+
+Each `TrainingRuntime` lives in a Kubeflow `Profile` namespace by default —
+change `metadata.namespace` to match the namespace you submit `TrainJob`s
+from. The YAMLs pin `:v0.1.0`; change the tag if you want to track a
+different release.
+
+## Submitting a TrainJob
+
+A shared smoke template applies to any of the runtimes — point
+`spec.runtimeRef.name` at the runtime you want to exercise:
+
+```bash
+kubectl apply -f https://raw.githubusercontent.com/alauda/aml-docs/master/docs/en/kubeflow/how_to/assets/training-runtimes/trainjob-smoke.yaml
+kubectl -n <your-namespace> get trainjobs
+kubectl -n <your-namespace> get pods -l trainer.kubeflow.org/trainjob-ancestor-step=trainer
+```
+
+Watch the trainer pod logs:
+
+```bash
+trainjob=$(kubectl -n <your-namespace> get trainjobs -o name | tail -1)
+kubectl -n <your-namespace> logs -f -l jobset.sigs.k8s.io/jobset-name=${trainjob##*/}-node
+```
+
+## Resource model on each device
+
+### NVIDIA GPU
+
+For clusters where each pod gets a whole device, request:
+
+```yaml
+resources:
+  limits:
+    nvidia.com/gpu: 1
+```
+
+For clusters running the HAMI vGPU scheduler, request a virtual slice
+instead:
+
+```yaml
+resources:
+  limits:
+    nvidia.com/gpualloc: 1      # one virtual GPU slot
+    nvidia.com/gpucores: 50     # 50% of one physical GPU's SM compute
+    nvidia.com/gpumem: "8192"   # 8 GiB of GPU memory
+```
+
+### Huawei Ascend NPU
+
+Set `runtimeClassName: ascend` on the pod template so the Ascend container
+runtime injects the driver libs and DCMI sockets.
+
+For clusters running the standard Huawei Kubernetes device-plugin, request:
+
+```yaml
+spec:
+  runtimeClassName: ascend
+  containers:
+    - resources:
+        limits:
+          huawei.com/Ascend910: "1"
+```
+
+For clusters running HAMI's vNPU scheduler — each Ascend 910B4 chip can be
+sliced into 20 cores / 32 GiB. Request a virtual slice via:
+
+```yaml
+spec:
+  schedulerName: hami-scheduler
+  runtimeClassName: ascend
+  containers:
+    - resources:
+        limits:
+          huawei.com/Ascend910B4: "1"          # one virtual NPU slot
+          huawei.com/Ascend910B4-memory: "8192" # 8 GiB of NPU HBM
+```
+
+With HAMI the node's `allocatable.huawei.com/Ascend910B4` reads `0` because
+HAMI manages allocation through its scheduler extender, not the standard
+device-plugin extended-resource path. If pods stay `Pending` with
+`hami-scheduler: 1 node unregistered`, verify the Ascend host driver is
+loaded (`/sys/bus/pci/drivers/davinci` exists, `npu-smi info` reports
+`Health: OK`) and that the node is labeled `ascend=on` so HAMI's
+device-plugin DaemonSet runs there.
+
+### Caveats by image
+
+- **`traininghub0.1-cu126-amd64`** — DeepSpeed JIT op compilation requires
+  `nvcc`. The base image ships CUDA *runtime* libs (via the official torch
+  cu126 wheels) but not the CUDA toolkit. For forward / backward / optimizer
+  with native torch ops the image works as-is; for DeepSpeed pre-compiled or
+  JIT-compiled ops, mount or install `nvidia-cuda-toolkit` and set
+  `CUDA_HOME` accordingly.
+- **`mindspeed-llm-cann8.5-arm64`** — `megatron.core` requires the legacy
+  `pkg_resources` API; if your job imports it, install
+  `setuptools<81` at the top of the entrypoint (or pin it in your derived
+  image). `import mindspeed_llm` currently fails on the
+  `megatron_adaptor_v2 → mindspeed.features_manager` chain because the
+  `mindspeed_llm` master branch expects a newer MindSpeed core than the
+  `core_r0.8.0` branch pinned in the image; the underlying
+  torch + torch_npu + megatron.core + mindspeed stack trains successfully on
+  the NPU without going through the MindSpeed-LLM adapter shim.
+
+## Building your own runtime
+
+The Containerfiles, the multi-arch buildkitd helper, an e2e harness, and
+the post-fix security scan evidence for every image are kept in the
+[`kubeflow-plugin/training-runtimes`](https://github.com/alauda/kubeflow-plugin/tree/main/training-runtimes)
+directory. Each framework image is a thin layer on top of one of the two
+base images (`torch2.6-cu126-amd64` or `torch2.6-cann8.5-arm64`), so
+deriving a new runtime is mostly
+`FROM docker.io/alaudadockerhub/torch2.6-cu126-amd64:v0.1.0` plus framework
+installs.