From 00529c5ca930868bbb54ae3978fd2de54715a42b Mon Sep 17 00:00:00 2001
From: selee <selee@inje-private.com>
Date: Tue, 10 Feb 2026 16:23:06 +0900
Subject: [PATCH] Comment out JAX runtime, update README with install results

- JAX runtime commented out (CRD does not support spec.mlPolicy.jax)
- Mark verification checklist items as completed
- Add known issues section (torchrun injection, JAX support)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 README.md                                     | 35 +++++++++----
 .../kubeflow-trainer/runtimes/runtimes.yaml   | 51 ++++++++++---------
 2 files changed, 52 insertions(+), 34 deletions(-)

diff --git a/README.md b/README.md
index 38dab78..303fc6e 100644
--- a/README.md
+++ b/README.md
@@ -130,6 +130,9 @@ helm install kubeflow-trainer charts/kubeflow-trainer-0.0.0-sha-48e7a93.tgz \
 kubectl apply -f manifests/kubeflow-trainer/runtimes/runtimes.yaml
 ```
 
+> **참고**: JAX runtime은 현재 CRD에서 `spec.mlPolicy.jax` 필드를 지원하지 않아 주석 처리되어 있음.
+> 설치되는 Runtime: deepspeed-distributed, mlx-distributed, torch-distributed, torchtune-llama3.2-1b/3b, torchtune-qwen2.5-1.5b
+
 ### 3.4 설치 확인
 
 ```bash
@@ -261,15 +264,29 @@ kubectl get pods -l trainer.kubeflow.org/trainjob-name=<trainjob-name> \
 
 ## 검증 체크리스트
 
-- [ ] Volcano Pod 3개 Running (`nodegroup=nd` 노드에 배치 확인)
-- [ ] Volcano CRD 생성됨 (jobs, podgroups, queues 등)
-- [ ] Kubeflow Trainer controller Running (`nodegroup=nd` 노드에 배치 확인)
-- [ ] JobSet controller Running (`nodegroup=nd` 노드에 배치 확인)
-- [ ] Kubeflow Trainer CRD 생성됨 (trainjobs, trainingruntimes 등)
-- [ ] ClusterTrainingRuntime 목록 확인 (deepspeed-distributed, torch-distributed 등)
-- [ ] Training Queue 생성됨
-- [ ] 테스트 TrainJob 제출 시 PodGroup 자동 생성 확인
-- [ ] 테스트 TrainJob Pod가 Volcano 스케줄러로 스케줄링됨
+- [x] Volcano Pod 3개 Running (`nodegroup=nd` 노드에 배치 확인)
+- [x] Volcano CRD 생성됨 (jobs, podgroups, queues 등)
+- [x] Kubeflow Trainer controller Running (`nodegroup=nd` 노드에 배치 확인)
+- [x] JobSet controller Running (`nodegroup=nd` 노드에 배치 확인)
+- [x] Kubeflow Trainer CRD 생성됨 (trainjobs, trainingruntimes 등)
+- [x] ClusterTrainingRuntime 목록 확인 (deepspeed-distributed, torch-distributed 등 6개)
+- [x] Training Queue 생성됨
+- [x] 테스트 TrainJob 제출 시 PodGroup 자동 생성 확인
+- [x] 테스트 TrainJob Pod가 Volcano 스케줄러로 스케줄링됨
+
+---
+
+## 알려진 이슈
+
+### Runtime에서 `command` 설정 시 torchrun 미주입
+
+Kubeflow Trainer torch runtime은 컨테이너에 `command`가 없을 때 자동으로 torchrun을 주입하여 `RANK`, `LOCAL_RANK`, `WORLD_SIZE` 등 환경변수를 설정합니다. Runtime에서 `command`를 직접 지정하면 torchrun 주입이 우회되어 분산 학습 환경변수가 설정되지 않습니다.
+
+현재 `volcano-trainjob-integration.yaml`의 Runtime에는 `command`가 설정되어 있으므로, 학습 스크립트 실행 방식을 조정해야 합니다.
+
+### JAX ClusterTrainingRuntime 미지원
+
+현재 CRD 버전에서 `spec.mlPolicy.jax` 필드를 지원하지 않아 JAX runtime은 주석 처리되어 있음.
 
 ---
 
diff --git a/manifests/kubeflow-trainer/runtimes/runtimes.yaml b/manifests/kubeflow-trainer/runtimes/runtimes.yaml
index be4d80f..a97b75d 100644
--- a/manifests/kubeflow-trainer/runtimes/runtimes.yaml
+++ b/manifests/kubeflow-trainer/runtimes/runtimes.yaml
@@ -54,31 +54,32 @@ spec:
         operator: All
         targetReplicatedJobs:
         - launcher
----
-apiVersion: trainer.kubeflow.org/v1alpha1
-kind: ClusterTrainingRuntime
-metadata:
-  labels:
-    trainer.kubeflow.org/framework: jax
-  name: jax-distributed
-spec:
-  mlPolicy:
-    jax: {}
-    numNodes: 1
-  template:
-    spec:
-      replicatedJobs:
-      - name: node
-        template:
-          metadata:
-            labels:
-              trainer.kubeflow.org/trainjob-ancestor-step: trainer
-          spec:
-            template:
-              spec:
-                containers:
-                - image: nvcr.io/nvidia/jax:25.10-py3
-                  name: node
+# ---
+# # JAX runtime - 현재 CRD에서 spec.mlPolicy.jax 미지원으로 주석처리
+# apiVersion: trainer.kubeflow.org/v1alpha1
+# kind: ClusterTrainingRuntime
+# metadata:
+#   labels:
+#     trainer.kubeflow.org/framework: jax
+#   name: jax-distributed
+# spec:
+#   mlPolicy:
+#     jax: {}
+#     numNodes: 1
+#   template:
+#     spec:
+#       replicatedJobs:
+#       - name: node
+#         template:
+#           metadata:
+#             labels:
+#               trainer.kubeflow.org/trainjob-ancestor-step: trainer
+#           spec:
+#             template:
+#               spec:
+#                 containers:
+#                 - image: nvcr.io/nvidia/jax:25.10-py3
+#                   name: node
 ---
 apiVersion: trainer.kubeflow.org/v1alpha1
 kind: ClusterTrainingRuntime