Merge branch 'verl-project:main' into main

Author: khazic

.github/workflows/e2e_ascend.yml

@@ -67,21 +67,16 @@ jobs:
   llm_rl_job:
     if: github.repository_owner == 'verl-project'
     name: E2E Ascend testing for RL training scenarios of LLM models
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 120
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env:
       HF_ENDPOINT: "https://hf-mirror.com"
       HF_HUB_ENABLE_HF_TRANSFER: "0" # This is more stable
     steps:
-      - name: Config third-party dependency download cache
-        run: |
-          sed -Ei 's@(ports|archive).ubuntu.com@cache-service.nginx-pypi-cache.svc.cluster.local:8081@g' /etc/apt/sources.list
-          pip config set global.index-url http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple
-          pip config set global.trusted-host cache-service.nginx-pypi-cache.svc.cluster.local
       - name: Check npu and CANN info
         run: |
           cat /usr/local/Ascend/ascend-toolkit/latest/"$(uname -i)"-linux/ascend_toolkit_install.info
@@ -132,21 +127,16 @@ jobs:
   vlm_rl_job:
     if: github.repository_owner == 'verl-project'
     name: E2E Ascend testing for RL training scenarios of VLM models
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 120
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env:
       HF_ENDPOINT: "https://hf-mirror.com"
       HF_HUB_ENABLE_HF_TRANSFER: "0" # This is more stable
     steps:
-      - name: Config third-party dependency download cache
-        run: |
-          sed -Ei 's@(ports|archive).ubuntu.com@cache-service.nginx-pypi-cache.svc.cluster.local:8081@g' /etc/apt/sources.list
-          pip config set global.index-url http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple
-          pip config set global.trusted-host cache-service.nginx-pypi-cache.svc.cluster.local
       - name: Check npu and CANN info
         run: |
           cat /usr/local/Ascend/ascend-toolkit/latest/"$(uname -i)"-linux/ascend_toolkit_install.info

.github/workflows/e2e_one_step_off_policy_ascend.yml

@@ -83,10 +83,10 @@ jobs:
   # Test FSDP2 strategy
   e2e_one_step_off_policy_fsdp2_ascend:
     if: github.repository_owner == 'verl-project'
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 60 # Increase this timeout value as needed
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env:
@@ -127,10 +127,10 @@ jobs:
   # Test Megatron strategy
   e2e_one_step_off_policy_megatron_ascend:
     if: github.repository_owner == 'verl-project'
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 60 # Increase this timeout value as needed
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env:

.github/workflows/e2e_ppo_trainer_megatron_vllm_2_ascend.yml

@@ -89,10 +89,10 @@ permissions:
 jobs:
   e2e_ppo_trainer_fsdp_vllm_ascend:
     if: github.repository_owner == 'verl-project'
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 90 # Increase this timeout value as needed
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env:
@@ -166,10 +166,10 @@ jobs:
 
   e2e_ppo_trainer_fsdp-qwen2_5vl-3b_ascend:
     if: github.repository_owner == 'verl-project'
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 60 # Increase this timeout value as needed
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env:

.github/workflows/e2e_sft_llm_ascend.yml

@@ -71,10 +71,10 @@ permissions:
 jobs:
   e2e_sft_llm_ascend:
     if: github.repository_owner == 'verl-project'
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 90 # Increase this timeout value as needed
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env:

.github/workflows/model_ascend.yml

@@ -63,10 +63,10 @@ permissions:
 jobs:
   model_rmpad_ascend:
     if: github.repository_owner == 'verl-project'
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 60 # Increase this timeout value as needed
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env:
@@ -112,10 +112,10 @@ jobs:
   # NOTE: List as an independent job to make rerun easier.
   model_rmpad_fsdp2_unstable_ascend:
     if: github.repository_owner == 'verl-project'
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 60
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env:

.github/workflows/nightly_ascend.yml

@@ -47,10 +47,10 @@ jobs:
   # Test ppo qwen3-8b fsdp+vllm
   nightlyCI_ppo-qwen3-8b-fsdp-vllm_ascend:
     if: github.repository_owner == 'verl-project'
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 180 # Increase this timeout value as needed
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env:
@@ -90,10 +90,10 @@ jobs:
   # Test grpo qwen25-7b-Instruct fsdp+vllm
   nightlyCI_grpo-qwen25-7b-Instruct-fsdp-vllm_ascend:
     if: github.repository_owner == 'verl-project'
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 180 # Increase this timeout value as needed
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env:
@@ -133,10 +133,10 @@ jobs:
   # Test grpo qwen25-vl-3b-Instruct fsdp+vllm
   nightlyCI_grpo-qwen25-vl-3b-Instruct-fsdp-vllm_ascend:
     if: github.repository_owner == 'verl-project'
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 180 # Increase this timeout value as needed
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env:

.github/workflows/npu_unit_tests.yml

@@ -73,10 +73,10 @@ permissions:
 jobs:
   npu_unit_tests:
     if: github.repository_owner == 'verl-project'
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 60 # Increase this timeout value as needed
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env:

.github/workflows/reward_model_vllm_ascend.yml

@@ -61,10 +61,10 @@ permissions:
 jobs:
   reward_model_vllm_ascend:
     if: github.repository_owner == 'verl-project'
-    runs-on: linux-aarch64-a2-8
+    runs-on: linux-aarch64-a2b3-8
     timeout-minutes: 60 # Increase this timeout value as needed
     container:
-      image: swr.ap-southeast-1.myhuaweicloud.com/base_image/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
+      image: swr.cn-southwest-2.myhuaweicloud.com/modelfoundry/ascend-ci/verl/verl:verl-8.5.0-910b-ubuntu22.04-py3.11-latest
       options: >-
         --shm-size 16g
     env: