Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
110 changes: 31 additions & 79 deletions .buildkite/pipeline_jax.yml
Original file line number Diff line number Diff line change
Expand Up @@ -19,34 +19,24 @@ steps:
- label: "E2E MLPerf tests for JAX models with quantization"
key: test_1
soft_fail: true
if: build.env("NIGHTLY") == "1"
env:
QUANTIZATION: "True"
agents:
queue: tpu_v6e_queue
commands:
- |
if [[ "$$NIGHTLY" == "1" ]]; then
.buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/mlperf.sh
else
echo "Skipping: NIGHTLY environment variable not set"
exit 0
fi
- .buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/mlperf.sh

- label: "E2E MLPerf tests for JAX new models"
key: test_2
soft_fail: true
if: build.env("NIGHTLY") == "1"
env:
NEW_MODEL_DESIGN: "1"
agents:
queue: tpu_v6e_queue
commands:
- |
if [[ "$$NIGHTLY" == "1" ]]; then
.buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/mlperf.sh
else
echo "Skipping: NIGHTLY environment variable not set"
exit 0
fi
- .buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/mlperf.sh

- label: "E2E MLPerf tests for JAX + vLLM models on single chip"
key: test_3
Expand All @@ -61,52 +51,35 @@ steps:
- label: "E2E MLperf tests for Llama4 models"
key: test_4
soft_fail: true
if: build.env("NIGHTLY") == "1"
env:
NEW_MODEL_DESIGN: "1"
USE_V6E8_QUEUE: "True"
agents:
queue: tpu_v6e_8_queue
commands:
- |
if [[ "$$NIGHTLY" == "1" ]]; then
.buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/mlperf.sh
else
echo "Skipping: NIGHTLY environment variable not set"
exit 0
fi

- .buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/mlperf.sh

- label: "E2E multi modality test"
key: test_5
soft_fail: true
if: build.env("NIGHTLY") == "1"
agents:
queue: tpu_v6e_queue
commands:
- |
if [[ "$$NIGHTLY" == "1" ]]; then
.buildkite/scripts/run_in_docker.sh \
bash -c 'python3 -m pytest -s -v -x /workspace/tpu_inference/tests/e2e/test_multi_modal_inference.py && \
bash /workspace/tpu_inference/tests/e2e/benchmarking/mm_bench.sh'
else
echo "Skipping: NIGHTLY environment variable not set"
exit 0
fi
- .buildkite/scripts/run_in_docker.sh \
bash -c 'python3 -m pytest -s -v -x /workspace/tpu_inference/tests/e2e/test_multi_modal_inference.py && \
bash /workspace/tpu_inference/tests/e2e/benchmarking/mm_bench.sh'

- label: "E2E speculative decoding test"
key: test_6
soft_fail: true
if: build.env("NIGHTLY") == "1"
agents:
queue: tpu_v6e_queue
commands:
- |
if [[ "$$NIGHTLY" == "1" ]]; then
.buildkite/scripts/run_in_docker.sh \
bash -c 'python3 -m pytest -s -v -x /workspace/tpu_inference/tests/e2e/test_speculative_decoding.py'
else
echo "Skipping: NIGHTLY environment variable not set"
exit 0
fi

- .buildkite/scripts/run_in_docker.sh \
bash -c 'python3 -m pytest -s -v -x /workspace/tpu_inference/tests/e2e/test_speculative_decoding.py'

- label: "JAX unit tests"
key: test_7
Expand Down Expand Up @@ -156,6 +129,8 @@ steps:
--ignore=/workspace/tpu_inference/tests/kernels/fused_moe_v1_test.py
else
echo "Skipping: no changes detected in kernels, tests/kernels, or requirements.txt"
buildkite-agent step update "label" "🚫 (Skipped) JAX unit tests - kernels"
buildkite-agent annotate "Step skipped because no changes were detected in kernel files." --style "info" --context "skip-kernels"
exit 0
fi

Expand All @@ -171,43 +146,36 @@ steps:
python3 -m pytest -s -v -x /workspace/tpu_inference/tests/kernels/collectives
else
echo "Skipping: no changes detected in kernels/collectives, tests/kernels/collectives, or requirements.txt"
buildkite-agent step update "label" "🚫 (Skipped) JAX unit tests - kernels"
buildkite-agent annotate "Step skipped because no changes were detected in kernel files." --style "info" --context "skip-kernels"
exit 0
fi

- label: "lora e2e tests for JAX + vLLM models single chip"
key: test_10
soft_fail: true
if: build.env("NIGHTLY") == "1"
agents:
queue: tpu_v6e_queue
commands:
- |
if [[ "$$NIGHTLY" == "1" ]]; then
.buildkite/scripts/run_in_docker.sh \
bash -c 'MODEL_IMPL_TYPE=vllm TPU_BACKEND_TYPE=jax python3 -m pytest -s -v -x /workspace/tpu_inference/tests/lora/test_lora.py'
else
echo "Skipping: NIGHTLY environment variable not set"
exit 0
fi
- .buildkite/scripts/run_in_docker.sh \
bash -c 'MODEL_IMPL_TYPE=vllm TPU_BACKEND_TYPE=jax python3 -m pytest -s -v -x /workspace/tpu_inference/tests/lora/test_lora.py'

- label: "E2E MLPerf tests for JAX + vLLM models on multiple chips"
key: test_11
soft_fail: true
if: build.env("NIGHTLY") == "1"
env:
MODEL_IMPL_TYPE: "vllm"
agents:
queue: tpu_v6e_8_queue
commands:
- |
if [[ "$$NIGHTLY" == "1" ]]; then
.buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/mlperf.sh
else
echo "Skipping: NIGHTLY environment variable not set"
exit 0
fi
- .buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/mlperf.sh

- label: "E2E MLperf tests for DeepSeek-R1 (no accuracy, 12-decoder layers only)"
key: test_12
soft_fail: true
if: build.env("NIGHTLY") == "1"
env:
NEW_MODEL_DESIGN: "1"
USE_V6E8_QUEUE: "True"
Expand All @@ -216,32 +184,20 @@ steps:
agents:
queue: tpu_v6e_8_queue
commands:
- |
if [[ "$$NIGHTLY" == "1" ]]; then
.buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/mlperf.sh -m deepseek-ai/DeepSeek-R1-0528 --use-dummy-weights
else
echo "Skipping: NIGHTLY environment variable not set"
exit 0
fi
- .buildkite/scripts/run_in_docker.sh bash /workspace/tpu_inference/tests/e2e/benchmarking/mlperf.sh -m deepseek-ai/DeepSeek-R1-0528 --use-dummy-weights

- label: "lora e2e tests for JAX + vLLM models multi chips"
key: test_13
soft_fail: true
if: build.env("NIGHTLY") == "1"
env:
USE_V6E8_QUEUE: "True"
VLLM_LOG_LEVEL: "INFO"
agents:
queue: tpu_v6e_8_queue
commands:
- |
if [[ "$$NIGHTLY" == "1" ]]; then
.buildkite/scripts/run_in_docker.sh \
bash -c 'MODEL_IMPL_TYPE=vllm TPU_BACKEND_TYPE=jax python3 -m pytest -s -v -x /workspace/tpu_inference/tests/lora/test_lora.py'
else
echo "Skipping: NIGHTLY environment variable not set"
exit 0
fi

- .buildkite/scripts/run_in_docker.sh \
bash -c 'MODEL_IMPL_TYPE=vllm TPU_BACKEND_TYPE=jax python3 -m pytest -s -v -x /workspace/tpu_inference/tests/lora/test_lora.py'

- label: "lora unit tests on single chip"
key: test_15
Expand All @@ -257,20 +213,16 @@ steps:
- label: "lora unit tests on multi chips"
key: test_16
soft_fail: true
if: build.env("NIGHTLY") == "1"
env:
USE_V6E8_QUEUE: "True"
VLLM_LOG_LEVEL: "INFO"
agents:
queue: tpu_v6e_8_queue
commands:
- |
if [[ "$$NIGHTLY" == "1" ]]; then
.buildkite/scripts/run_in_docker.sh \
bash -c 'python3 -m pytest -s -v -x /workspace/tpu_inference/tests/lora/test_layers.py'
else
echo "Skipping: NIGHTLY environment variable not set"
exit 0
fi
- .buildkite/scripts/run_in_docker.sh \
bash -c 'python3 -m pytest -s -v -x /workspace/tpu_inference/tests/lora/test_layers.py'

# -----------------------------------------------------------------
# NOTIFICATION STEP
# -----------------------------------------------------------------
Expand Down
4 changes: 4 additions & 0 deletions .buildkite/scripts/check_results.sh
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,10 @@ echo "--- Checking Test Outcomes"

for KEY in "$@"; do
OUTCOME=$(buildkite-agent step get "outcome" --step "${KEY}" || echo "skipped")
if [ -z "$OUTCOME" ]; then
OUTCOME="skipped"
fi

echo "Step ${KEY} outcome: ${OUTCOME}"

if [ "${OUTCOME}" != "passed" ] && [ "${OUTCOME}" != "skipped" ] ; then
Expand Down