From 40cbd8ef7ff2d04dcf2ab0672b7e19a56303bfa9 Mon Sep 17 00:00:00 2001 From: dongbo910220 <32610838+dongbo910220@users.noreply.github.com> Date: Tue, 27 Jan 2026 23:55:05 +0800 Subject: [PATCH] [Fix] make images LoRA e2e less flaky (#978) Signed-off-by: dongbo910220 <1275604947@qq.com> Co-authored-by: Hongsheng Liu --- .buildkite/pipeline.yml | 17 +++++++++++++++++ .../test_images_generations_lora.py | 10 +++++----- 2 files changed, 22 insertions(+), 5 deletions(-) diff --git a/.buildkite/pipeline.yml b/.buildkite/pipeline.yml index dcc9b3064..d1e71aba2 100644 --- a/.buildkite/pipeline.yml +++ b/.buildkite/pipeline.yml @@ -41,6 +41,23 @@ steps: volumes: - "/fsx/hf_cache:/fsx/hf_cache" + - label: "Diffusion Images API LoRA E2E" + timeout_in_minutes: 20 + depends_on: image-build + commands: + - pytest -s -v tests/e2e/online_serving/test_images_generations_lora.py + agents: + queue: "gpu_1_queue" # g6.4xlarge instance on AWS, has 1 L4 GPU + plugins: + - docker#v5.2.0: + image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT + always-pull: true + propagate-environment: true + environment: + - "HF_HOME=/fsx/hf_cache" + volumes: + - "/fsx/hf_cache:/fsx/hf_cache" + - label: "Diffusion Model CPU offloading Test" timeout_in_minutes: 20 depends_on: image-build diff --git a/tests/e2e/online_serving/test_images_generations_lora.py b/tests/e2e/online_serving/test_images_generations_lora.py index e5fabcb81..82044cd52 100644 --- a/tests/e2e/online_serving/test_images_generations_lora.py +++ b/tests/e2e/online_serving/test_images_generations_lora.py @@ -224,11 +224,11 @@ def test_images_generations_per_request_lora_switching(omni_server: OmniServer, base_img = _post_images(omni_server, _basic_payload()) base_slice = _image_blue_tail_slice(base_img) - # Adapter A: apply delta to Q slice only. + # Adapter A: apply delta to V slice only. lora_a_dir = tmp_path / "zimage_lora_a" - _write_zimage_lora(lora_a_dir, q_scale=0.1) + _write_zimage_lora(lora_a_dir, v_scale=8.0) payload_a = _basic_payload() - payload_a["lora"] = {"name": "a", "path": str(lora_a_dir), "scale": 2.0} + payload_a["lora"] = {"name": "a", "path": str(lora_a_dir), "scale": 64.0} img_a = _post_images(omni_server, payload_a) a_slice = _image_blue_tail_slice(img_a) _assert_slice_diff(a_slice, base_slice, label="lora_a_vs_base") @@ -236,9 +236,9 @@ def test_images_generations_per_request_lora_switching(omni_server: OmniServer, # Adapter B: apply delta to K slice only (should differ from adapter A). lora_b_dir = tmp_path / "zimage_lora_b" - _write_zimage_lora(lora_b_dir, k_scale=0.1) + _write_zimage_lora(lora_b_dir, k_scale=4.0) payload_b = _basic_payload() - payload_b["lora"] = {"name": "b", "path": str(lora_b_dir), "scale": 2.0} + payload_b["lora"] = {"name": "b", "path": str(lora_b_dir), "scale": 64.0} img_b = _post_images(omni_server, payload_b) b_slice = _image_blue_tail_slice(img_b) _assert_slice_diff(b_slice, base_slice, label="lora_b_vs_base")