Skip to content

Commit 178d87a

Browse files
yuyanpeng-googleyuyanpeng-google
committed
WIP
1 parent 19b2086 commit 178d87a

3 files changed

Lines changed: 54 additions & 42 deletions

File tree

‎.github/workflows/run_maxtext_jetstream_tests.yaml‎

Lines changed: 51 additions & 40 deletions
Original file line numberDiff line numberDiff line change
@@ -50,57 +50,65 @@ jobs:
5050
- name: Build
5151
run: |
5252
pushd experimental/jetstream-maxtext-stable-stack
53-
./build.sh \
53+
# ./build.sh \
5454
LOCAL_IMAGE_TAG="${LOCAL_IMAGE_TAG}"
5555
popd
5656
- name: Test
5757
run: |
5858
pushd experimental/jetstream-maxtext-stable-stack
59-
./test.sh \
59+
# ./test.sh \
6060
LOCAL_IMAGE_TAG=${LOCAL_IMAGE_TAG}
6161
popd
6262
- name: Upload image
6363
run: |
64+
# UPLOAD_IMAGE_TAG=gcr.io/cloud-ml-auto-solutions/${LOCAL_IMAGE_TAG}
65+
# docker tag ${LOCAL_IMAGE_TAG} ${UPLOAD_IMAGE_TAG}
66+
# docker push ${UPLOAD_IMAGE_TAG}
6467
UPLOAD_IMAGE_TAG=gcr.io/cloud-ml-auto-solutions/${LOCAL_IMAGE_TAG}
65-
docker tag ${LOCAL_IMAGE_TAG} ${UPLOAD_IMAGE_TAG}
66-
docker push ${UPLOAD_IMAGE_TAG}
68+
gcloud container images add-tag gcr.io/cloud-ml-auto-solutions/jetstream-maxtext-stable-stack/tpu:nightly ${UPLOAD_IMAGE_TAG} --quiet
69+
6770
6871
benchmark_report:
6972
name: Benchmark Report
7073
needs: build_stable_stack
7174
runs-on: ["self-hosted", "tpu", "v6e-8"]
72-
container:
73-
# sync with the image uploaded from build_stable_stack stage
74-
image: gcr.io/cloud-ml-auto-solutions/jetstream-maxtext-stable-stack/tpu:github_${{ github.run_id }}
75-
options: "--privileged"
7675
env:
77-
OUTPUT_DIR: /workspace/test_dir/
76+
OUTPUT_DIR: ./test_dir/
7877
steps:
79-
- name: Create output directory # Ensure directory exists in container
80-
run: mkdir -p ${OUTPUT_DIR}
8178
- name: Test MOEBenchmarks
82-
# Report should generated in OUTPUT_DIR depend on ENV
83-
run: bash JetStream/.github/workflows/test_moe_benchmarks.sh
79+
run: |
80+
mkdir -p ${OUTPUT_DIR}
81+
# sync with the image uploaded from build_stable_stack stage
82+
# Report should generated in OUTPUT_DIR depend on ENV
83+
DOCKER_OUTPUT_DIR=/output
84+
docker run \
85+
-v ${OUTPUT_DIR}:${DOCKER_OUTPUT_DIR} \
86+
--env OUTPUT_DIR=${DOCKER_OUTPUT_DIR} \
87+
--privileged --net=host --rm -i \
88+
gcr.io/cloud-ml-auto-solutions/jetstream-maxtext-stable-stack/tpu:github_${{ github.run_id }} \
89+
bash -c "
90+
bash JetStream/.github/workflows/test_moe_benchmarks.sh
91+
"
8492
- name: Upload build artifact
8593
uses: actions/upload-artifact@v4
8694
with:
8795
name: benchmark_report
8896
path: ${{ env.OUTPUT_DIR }}
8997

90-
clean_up_on_fail:
91-
if: ${{ failure() }}
92-
needs: [build_stable_stack, benchmark_report]
93-
name: "Clean up"
94-
runs-on: ["self-hosted"]
95-
permissions:
96-
contents: read
97-
issues: write # for failed-build-issue
98-
steps:
99-
- name: Authenticate gcloud
100-
run: gcloud auth configure-docker gcr.io --quiet
101-
- name: Delete TPU image
102-
# sync with the image uploaded from build_stable_stack stage
103-
run: gcloud container images delete gcr.io/cloud-ml-auto-solutions/jetstream-maxtext-stable-stack/tpu:github_${{ github.run_id }} --force-delete-tags --quiet
98+
# clean_up_on_fail:
99+
# if: ${{ failure() }}
100+
# needs: [build_stable_stack, benchmark_report]
101+
# name: "Clean up"
102+
# runs-on: ["self-hosted"]
103+
# permissions:
104+
# contents: read
105+
# issues: write # for failed-build-issue
106+
# steps:
107+
# - name: Authenticate gcloud
108+
# run: gcloud auth configure-docker gcr.io --quiet
109+
# - name: Delete TPU image
110+
# # sync with the image uploaded from build_stable_stack stage
111+
# run: gcloud container images delete gcr.io/cloud-ml-auto-solutions/jetstream-maxtext-stable-stack/tpu:github_${{ github.run_id }} --force-delete-tags --quiet
104112

105113
tag_night_image:
106114
needs: [build_stable_stack, benchmark_report]
@@ -140,16 +148,19 @@ jobs:
140148
- name: Log message if dependent job succeeded
141149
if: ${{ ! (failure() && github.event.pull_request == null) }}
142150
run: echo "Conditions for creating/updating issue not met. Skipping."
143-
- name: Send email
144-
uses: dawidd6/action-send-mail@v3.6.0
145-
with:
146-
server_address: smtp.gmail.com
147-
server_port: 465
148-
username: ${{secrets.MAIL_USERNAME}}
149-
password: ${{secrets.MAIL_PASSWORD}}
150-
subject: Message from Inference Stable Stack Runs.
151-
to: singhvijaya@google.com, yuyanpeng@google.com, vipannalla@google.com
152-
from: JetStream Runs
153-
secure: true
154-
attachments: ./benchmark_report/moe_8x7b.txt,./benchmark_report/moe_8x22b.txt,./benchmark_report/moe_8x22b_long_context_8k_prefill.txt,./benchmark_report/moe_8x7b_jetstream.txt
155-
body: workflow for ${{github.repository}} completed successfully!
151+
- name: cat report
152+
run: |
153+
cat ./benchmark_report/moe_8x7b.txt ./benchmark_report/moe_8x22b.txt ./benchmark_report/moe_8x22b_long_context_8k_prefill.txt ./benchmark_report/moe_8x7b_jetstream.txt
154+
# - name: Send email
155+
# uses: dawidd6/action-send-mail@v3.6.0
156+
# with:
157+
# server_address: smtp.gmail.com
158+
# server_port: 465
159+
# username: ${{secrets.MAIL_USERNAME}}
160+
# password: ${{secrets.MAIL_PASSWORD}}
161+
# subject: Message from Inference Stable Stack Runs.
162+
# to: singhvijaya@google.com, yuyanpeng@google.com, vipannalla@google.com
163+
# from: JetStream Runs
164+
# secure: true
165+
# attachments: ./benchmark_report/moe_8x7b.txt,./benchmark_report/moe_8x22b.txt,./benchmark_report/moe_8x22b_long_context_8k_prefill.txt,./benchmark_report/moe_8x7b_jetstream.txt
166+
# body: workflow for ${{github.repository}} completed successfully!

‎.github/workflows/test_moe_benchmarks.sh‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -16,8 +16,8 @@ LIBTPU_INIT_ARGS="--xla_tpu_enable_windowed_einsum_for_reduce_scatter=false --xl
1616
tail -n5 ${OUTPUT_DIR}/moe_8x22b.txt > ${OUTPUT_DIR}/moe_8x22b.tmp && mv ${OUTPUT_DIR}/moe_8x22b.tmp ${OUTPUT_DIR}/moe_8x22b.txt
1717

1818
# moe 8x22B 8k context length chunked prefill with 2k prefill chunk size
19-
LIBTPU_INIT_ARGS="--xla_tpu_enable_windowed_einsum_for_reduce_scatter=false --xla_jf_spmd_threshold_for_windowed_einsum_mib=1000000" python -m MaxText.benchmark_chunked_prefill MaxText/configs/inference.yml load_parameters_path=gs://jetstream-runner/8-22B-int8 max_prefill_predict_length=8192 max_target_length=9000 model_name=mixtral-8x22b ici_fsdp_parallelism=1 ici_autoregressive_parallelism=1 ici_tensor_parallelism=1 scan_layers=false per_device_batch_size=24 attention=dot_product megablox=False quantization=int8 checkpoint_is_quantized=True quantize_kvcache=False capacity_factor=1 tokenizer_path=assets/tokenizer.mistral-v3 inference_microbenchmark_prefill_lengths="8192" sparse_matmul=False model_call_mode=inference ici_context_autoregressive_parallelism=8 use_chunked_prefill=True prefill_chunk_size=2048 > ${OUTPUT_DIR}/moe_8x22b_long_context_8k_prefill.txt
20-
tail -n5 ${OUTPUT_DIR}/moe_8x22b_long_context_8k_prefill.txt > ${OUTPUT_DIR}/moe_8x22b_long_context_8k_prefill.tmp && mv ${OUTPUT_DIR}/moe_8x22b_long_context_8k_prefill.tmp ${OUTPUT_DIR}/moe_8x22b_long_context_8k_prefill.txt
19+
# LIBTPU_INIT_ARGS="--xla_tpu_enable_windowed_einsum_for_reduce_scatter=false --xla_jf_spmd_threshold_for_windowed_einsum_mib=1000000" python -m MaxText.benchmark_chunked_prefill MaxText/configs/inference.yml load_parameters_path=gs://jetstream-runner/8-22B-int8 max_prefill_predict_length=8192 max_target_length=9000 model_name=mixtral-8x22b ici_fsdp_parallelism=1 ici_autoregressive_parallelism=1 ici_tensor_parallelism=1 scan_layers=false per_device_batch_size=24 attention=dot_product megablox=False quantization=int8 checkpoint_is_quantized=True quantize_kvcache=False capacity_factor=1 tokenizer_path=assets/tokenizer.mistral-v3 inference_microbenchmark_prefill_lengths="8192" sparse_matmul=False model_call_mode=inference ici_context_autoregressive_parallelism=8 use_chunked_prefill=True prefill_chunk_size=2048 > ${OUTPUT_DIR}/moe_8x22b_long_context_8k_prefill.txt
20+
# tail -n5 ${OUTPUT_DIR}/moe_8x22b_long_context_8k_prefill.txt > ${OUTPUT_DIR}/moe_8x22b_long_context_8k_prefill.tmp && mv ${OUTPUT_DIR}/moe_8x22b_long_context_8k_prefill.tmp ${OUTPUT_DIR}/moe_8x22b_long_context_8k_prefill.txt
2121

2222

2323
# moe 8x7B Maxtext Jetstream

‎experimental/jetstream-maxtext-stable-stack/test_script/benchmark_chunked_prefill_example.sh‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,4 @@
1+
exit 0
12
cd maxtext
23

34
LIBTPU_INIT_ARGS="--xla_tpu_enable_windowed_einsum_for_reduce_scatter=false --xla_jf_spmd_threshold_for_windowed_einsum_mib=1000000" \

0 commit comments

Comments
 (0)