A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 18:30 UTC
[V1] TPU - CI/CD use smaller model (#15054)
Signed-off-by: Alexander Matveev <amatveev@redhat.com>
Alexander Matveev committed
Mar 18, 2025 at 17:39 UTC
72a8639b68964ba50a019856f2fabd3c4fdbaa3f
2 files changed
+10
-9
.buildkite/run-tpu-v1-test.sh
+7
-7
@@ -20,17 +20,17 @@ docker run --privileged --net host --shm-size=16G -it \
20
&& python3 -m pip install pytest \
21
&& python3 -m pip install lm_eval[api]==0.4.4 \
22
&& echo TEST_1 \
23
- && VLLM_USE_V1=1 pytest -v -s /workspace/vllm/tests/v1/tpu/test_basic.py \
23
+ && VLLM_USE_V1=1 python3 /workspace/vllm/tests/tpu/test_compilation.py \
24
&& echo TEST_2 \
25
- && VLLM_USE_V1=1 pytest -v -s /workspace/vllm/tests/entrypoints/llm/test_accuracy.py::test_lm_eval_accuracy_v1_engine \
25
+ && VLLM_USE_V1=1 pytest -v -s /workspace/vllm/tests/v1/tpu/test_basic.py \
26
&& echo TEST_3 \
27
- && VLLM_USE_V1=1 pytest -s -v /workspace/vllm/tests/tpu/test_quantization_accuracy.py \
27
+ && VLLM_USE_V1=1 pytest -v -s /workspace/vllm/tests/entrypoints/llm/test_accuracy.py::test_lm_eval_accuracy_v1_engine \
28
&& echo TEST_4 \
29
- && VLLM_USE_V1=1 python3 /workspace/vllm/examples/offline_inference/tpu.py" \
29
+ && VLLM_USE_V1=1 pytest -s -v /workspace/vllm/tests/tpu/test_quantization_accuracy.py \
30
&& echo TEST_5 \
31
- && VLLM_USE_V1=1 python3 /workspace/vllm/tests/tpu/test_compilation.py \
32
-
31
+ && VLLM_USE_V1=1 python3 /workspace/vllm/examples/offline_inference/tpu.py" \
32
+
33
34
-# TODO: Fix these tests
34
+# TODO: This test fails because it uses RANDOM_SEED sampling
35
# && VLLM_USE_V1=1 pytest -v -s /workspace/vllm/tests/tpu/test_custom_dispatcher.py \
36
tests/v1/tpu/test_basic.py
+3
-2
@@ -15,9 +15,10 @@ if TYPE_CHECKING:
15
from tests.conftest import VllmRunner
16
17
MODELS = [
18
+ "Qwen/Qwen2.5-1.5B-Instruct",
19
+ # TODO: Enable this models with v6e
20
# "Qwen/Qwen2-7B-Instruct",
19
- "meta-llama/Llama-3.1-8B",
20
- # TODO: Add models here as necessary
21
+ # "meta-llama/Llama-3.1-8B",
22
]
23
24
TENSOR_PARALLEL_SIZES = [1]