2222 delete :
2323
2424jobs :
25- prepare_cache :
26- name : Prepare testmon cache
27- if : |
28- github.event_name == 'create' &&
29- github.event.ref_type == 'branch' &&
30- github.event.repository.full_name == 'hpcaitech/ColossalAI'
31- runs-on : [self-hosted, gpu]
32- container :
33- image : hpcaitech/pytorch-cuda:1.12.0-11.3.0
34- options : --rm
35- timeout-minutes : 5
36- defaults :
37- run :
38- shell : bash
39- steps :
40- - name : Copy testmon cache
41- run : | # branch name may contain slash, we need to replace it with space
42- export REF_BRANCH=$(echo ${{ github.event.ref }} | sed "s/\// /")
43- if [ -d /github/home/testmon_cache/${MAIN_BRANCH} ]; then
44- cp -p -r /github/home/testmon_cache/${MAIN_BRANCH} "/github/home/testmon_cache/${REF_BRANCH}"
45- fi
46- env :
47- MAIN_BRANCH : ${{ github.event.master_branch }}
48-
49- prepare_cache_for_pr :
50- name : Prepare testmon cache for PR
51- if : |
52- github.event_name == 'pull_request' &&
53- (github.event.action == 'opened' || github.event.action == 'reopened' || (github.event.action == 'edited' && github.event.changes.base != null)) &&
54- github.event.pull_request.base.repo.full_name == 'hpcaitech/ColossalAI'
55- runs-on : [self-hosted, gpu]
56- container :
57- image : hpcaitech/pytorch-cuda:1.12.0-11.3.0
58- options : --rm
59- timeout-minutes : 5
60- defaults :
61- run :
62- shell : bash
63- concurrency :
64- group : ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-repare-cache
65- cancel-in-progress : true
66- steps :
67- - name : Copy testmon cache
68- run : | # branch name may contain slash, we need to replace it with space
69- export BASE=$(echo ${{ github.event.pull_request.base.ref }} | sed "s/\// /")
70- if [ -d "/github/home/testmon_cache/${BASE}" ] && [ ! -z "$(ls -A "/github/home/testmon_cache/${BASE}")" ]; then
71- mkdir -p /github/home/testmon_cache/_pull/${PR_NUMBER} && cp -p -r "/github/home/testmon_cache/${BASE}"/.testmondata* /github/home/testmon_cache/_pull/${PR_NUMBER}
72- fi
73- env :
74- PR_NUMBER : ${{ github.event.number }}
75-
7625 detect :
7726 name : Detect file change
7827 if : |
14089 if : needs.detect.outputs.anyLibraryFileChanged == 'true'
14190 runs-on : [self-hosted, gpu]
14291 container :
143- image : hpcaitech/pytorch-cuda:1.12.0-11.3 .0
144- options : --gpus all --rm -v /data/scratch/cifar-10:/data/scratch/cifar-10 -v /data/scratch/llama-tiny:/data/scratch/llama-tiny
92+ image : hpcaitech/pytorch-cuda:2.1.0-12.1 .0
93+ options : --gpus all --rm -v /dev/shm -v /data/scratch/llama-tiny:/data/scratch/llama-tiny
14594 timeout-minutes : 60
14695 defaults :
14796 run :
@@ -168,12 +117,13 @@ jobs:
168117 cd TensorNVMe
169118 conda install cmake
170119 pip install -r requirements.txt
171- pip install -v .
120+ DISABLE_URING=1 pip install -v .
172121
173122name : Store TensorNVMe Cache
174123 run : |
175124 cd TensorNVMe
176125 cp -p -r ./build /github/home/tensornvme_cache/
126+ cp -p -r ./cmake-build /github/home/tensornvme_cache/
177127
178128name : Checkout Colossal-AI
179129 uses : actions/checkout@v2
@@ -190,39 +140,32 @@ jobs:
190140
191141name : Install Colossal-AI
192142 run : |
193- CUDA_EXT =1 pip install -v -e .
143+ BUILD_EXT =1 pip install -v -e .
194144 pip install -r requirements/requirements-test.txt
195145
196146name : Store Colossal-AI Cache
197147 run : |
198148 # -p flag is required to preserve the file timestamp to avoid ninja rebuild
199149 cp -p -r /__w/ColossalAI/ColossalAI/build /github/home/cuda_ext_cache/
200150
201- name : Restore Testmon Cache
202- run : |
203- if [ -d /github/home/testmon_cache/_pull/${PR_NUMBER} ] && [ ! -z "$(ls -A /github/home/testmon_cache/_pull/${PR_NUMBER})" ]; then
204- cp -p -r /github/home/testmon_cache/_pull/${PR_NUMBER}/.testmondata* /__w/ColossalAI/ColossalAI/
205- fi
206- env :
207- PR_NUMBER : ${{ github.event.number }}
208-
209151name : Execute Unit Testing
210152 run : |
211- CURL_CA_BUNDLE="" PYTHONPATH=$PWD pytest -m "not largedist" --testmon --testmon-forceselect --testmon-cov=. --durations=10 tests/
153+ CURL_CA_BUNDLE="" PYTHONPATH=$PWD FAST_TEST=1 pytest \
154+ -m "not largedist" \
155+ --durations=0 \
156+ --ignore tests/test_analyzer \
157+ --ignore tests/test_auto_parallel \
158+ --ignore tests/test_fx \
159+ --ignore tests/test_autochunk \
160+ --ignore tests/test_gptq \
161+ --ignore tests/test_infer_ops \
162+ --ignore tests/test_legacy \
163+ --ignore tests/test_smoothquant \
164+ tests/
212165env :
213- DATA : /data/scratch/cifar-10
214- NCCL_SHM_DISABLE : 1
215166 LD_LIBRARY_PATH : /github/home/.tensornvme/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64
216- TESTMON_CORE_PKGS : /__w/ColossalAI/ColossalAI/requirements/requirements.txt,/__w/ColossalAI/ColossalAI/requirements/requirements-test.txt
217167 LLAMA_PATH : /data/scratch/llama-tiny
218168
219- - name : Store Testmon Cache
220- run : |
221- mkdir -p /github/home/testmon_cache/_pull/${PR_NUMBER}
222- cp -p -r /__w/ColossalAI/ColossalAI/.testmondata* /github/home/testmon_cache/_pull/${PR_NUMBER}/
223- env :
224- PR_NUMBER : ${{ github.event.number }}
225-
226169 - name : Collate artifact
227170 env :
228171 PR_NUMBER : ${{ github.event.number }}
@@ -259,54 +202,3 @@ jobs:
259202 with :
260203 name : report
261204 path : report/
262-
263- store_cache :
264- name : Store testmon cache for PR
265- if : |
266- github.event_name == 'pull_request' &&
267- github.event.action == 'closed' &&
268- github.event.pull_request.base.repo.full_name == 'hpcaitech/ColossalAI'
269- runs-on : [self-hosted, gpu]
270- container :
271- image : hpcaitech/pytorch-cuda:1.12.0-11.3.0
272- options : --rm
273- timeout-minutes : 5
274- defaults :
275- run :
276- shell : bash
277- steps :
278- - name : Store testmon cache if possible
279- if : github.event.pull_request.merged == true
280- run : | # branch name may contain slash, we need to replace it with space
281- export BASE=$(echo ${{ github.event.pull_request.base.ref }} | sed "s/\// /")
282- if [ -d /github/home/testmon_cache/_pull/${PR_NUMBER} ] && [ ! -z "$(ls -A /github/home/testmon_cache/_pull/${PR_NUMBER})" ]; then
283- cp -p -r /github/home/testmon_cache/_pull/${PR_NUMBER}/.testmondata* "/github/home/testmon_cache/${BASE}/"
284- fi
285- env :
286- PR_NUMBER : ${{ github.event.pull_request.number }}
287-
288- - name : Remove testmon cache
289- run : |
290- rm -rf /github/home/testmon_cache/_pull/${PR_NUMBER}
291- env :
292- PR_NUMBER : ${{ github.event.pull_request.number }}
293-
294- remove_cache :
295- name : Remove testmon cache
296- if : |
297- github.event_name == 'delete' &&
298- github.event.ref_type == 'branch' &&
299- github.event.repository.full_name == 'hpcaitech/ColossalAI'
300- runs-on : [self-hosted, gpu]
301- container :
302- image : hpcaitech/pytorch-cuda:1.12.0-11.3.0
303- options : --rm
304- timeout-minutes : 5
305- defaults :
306- run :
307- shell : bash
308- steps :
309- - name : Remove testmon cache
310- run : | # branch name may contain slash, we need to replace it with space
311- export BASE=$(echo ${{ github.event.ref }} | sed "s/\// /")
312- rm -rf "/github/home/testmon_cache/${BASE}"
0 commit comments