Skip to content

Commit a8d4b8a

Browse files
authored
fix: nvidia container/mthreads env/huawei env for weekly ci job (flagos-ai#4859)
1 parent e12a220 commit a8d4b8a

3 files changed

Lines changed: 16 additions & 11 deletions

File tree

.github/configs/weekly/H800.yml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -11,7 +11,7 @@ runner_labels: h800-flaggems-weekly
1111
use_container: true
1212

1313
# Container image for this backends
14-
container_image: harbor.baai.ac.cn/flaggems/nvidia-flaggems-test-h800:20260701
14+
container_image: harbor.baai.ac.cn/flaggems/nvidia-flaggems-test-h800-flagtree0.6.0:202607171136
1515

1616
# Container volumes to mount for this backends
1717
container_volumes:

.github/configs/weekly/MTT-S5000.yml

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -40,6 +40,7 @@ test_gpus: "0,1,2,3,4,5,6,7"
4040
# Extra environment variables injected only into the "Run weekly tests" step
4141
test_env:
4242
PIP_CACHE_DIR: /root/.cache/pip
43+
FLAGTREE_BACKEND: mthreads
4344

4445
# GPU availability check script for this backend
4546
gpu_check_script: tools/gpu_check_mthreads.sh

.github/workflows/weekly.yaml

Lines changed: 14 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -15,8 +15,8 @@ on:
1515
workflow_dispatch:
1616

1717
schedule:
18-
# Monday 02:00 Beijing time = Sunday 18:00 UTC
19-
- cron: '0 18 * * 0'
18+
# Monday 00:10 Beijing time = Sunday 16:10 UTC
19+
- cron: '10 16 * * 0'
2020

2121
defaults:
2222
run:
@@ -116,11 +116,20 @@ jobs:
116116
# that prevent conda init / backend env setup from running in
117117
# non-interactive CI shells. Extract those lines (conda init,
118118
# `source .../set_env.sh`, FLAGTREE_BACKEND export, and the
119-
# `. $HOME/.local/bin/env` sourcing) and append to .bash_profile
119+
# `. /root/.local/bin/env` sourcing) and append to .bash_profile
120120
# instead, which is sourced by login shells without any
121121
# interactive check.
122+
#
123+
# huawei: set_env.sh
124+
# mthreads: $HOME/.local/bin/env
125+
# kunlunxin: conda.sh and conda activate
126+
#
122127
if [ -f /root/.bashrc ]; then
123-
grep -E '(conda\.sh|conda activate|set_env\.sh|FLAGTREE_BACKEND|\.local/bin/env)' /root/.bashrc >> "$HOME/.bash_profile" || true
128+
# Pin $HOME -> /root: extracted lines use "$HOME/..." (e.g.
129+
# . "$HOME/.local/bin/env"), but the login shell that sources
130+
# .bash_profile has $HOME set to the GHA home, not /root.
131+
grep -E '(conda\.sh|conda activate|set_env\.sh|\.local/bin/env)' /root/.bashrc \
132+
| sed 's#\$HOME\/\.local#/root\/\.local#g' >> "$HOME/.bash_profile" || true
124133
fi
125134
126135
- name: Install FlagGems dependencies
@@ -138,8 +147,6 @@ jobs:
138147
# env: ${{ fromJSON(matrix.test_env) }}
139148
# shell: bash --login -e -o pipefail {0}
140149
# run: |
141-
# # remove cache to avoid potential issues with stale files
142-
# rm -rf ~/.triton/cache && rm -rf ~/.flaggems
143150
# # run tests with specified GPUs and output directory
144151
# printenv | grep -viE '^(https?_proxy|all_proxy)=' || true
145152
# python3 tools/run_tests.py \
@@ -159,8 +166,6 @@ jobs:
159166
env: ${{ fromJSON(matrix.test_env) }}
160167
shell: bash --login -e -o pipefail {0}
161168
run: |
162-
# remove cache to avoid potential issues with stale files
163-
rm -rf ~/.triton/cache && rm -rf ~/.flaggems
164169
# run tests with specified GPUs and output directory
165170
# Print env for debugging; drop proxy vars (http(s)_proxy/all_proxy,
166171
# any case) since they may embed user:password that GitHub Actions
@@ -177,7 +182,6 @@ jobs:
177182
python3 tools/psum_html -o ${OUTPUT_DIR}/result.html ${OUTPUT_DIR}
178183
179184
- name: Package results
180-
id: package-results
181185
run: |
182186
ARCHIVE="${{ runner.temp }}/weekly-${{ matrix.name }}-${DATE}.tar.gz"
183187
tar -czf "${ARCHIVE}" -C "$(dirname "${OUTPUT_DIR}")" "$(basename "${OUTPUT_DIR}")"
@@ -264,7 +268,7 @@ jobs:
264268
# timeout-minutes: 720
265269
# env: ${{ fromJSON(matrix.test_env) }}
266270
# run: |
267-
# # remove cache to avoid potential issues with stale files
271+
# #remove cache to avoid potential issues with stale files
268272
# rm -rf ~/.triton/cache && rm -rf ~/.flaggems
269273
# # run tests with specified GPUs and output directory
270274
# python3 tools/run_tests.py \

0 commit comments

Comments
 (0)