From 1f0e6fd2f3349d492f7bf9c54320b3cd4840f19d Mon Sep 17 00:00:00 2001 From: Colin Swaney Date: Thu, 27 Aug 2026 20:33:53 -0400 Subject: [PATCH] chore(lib): bump vllm/vllm-openai to v0.26.0 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Two-year jump from v0.20.0 (Aug 2024) → v0.26.0 (Jul 2026). The default launch template's flags — --model, --port, --revision, --trust-remote-code, --tensor-parallel-size — have not moved between the two versions, and the Disable Thinking toggle's --default-chat-template-kwargs shape is unchanged. Motivation: v0.20 pre-dates a lot of the model architectures we now cache, notably Gemma 4 (v0.21+), gpt-oss (v0.24), and TranslateGemma (v0.26). We already confirmed Gemma-4-12b-it, Gemma-3-4b-it, and DeepSeek-OCR-2 launch cleanly on v0.26.0 via a runtime env override. Follow the earlier tigerflow-ml pattern for the test that used to hard-code the default tag: assert against `DEFAULT_IMAGES` instead of the literal so future bumps don't touch this file. Refs #512 --- lib/src/blackfish/cli/__main__.py | 2 +- lib/src/blackfish/client.py | 2 +- lib/src/blackfish/server/images.py | 2 +- lib/tests/cli/test_cli_run.py | 4 +++- 4 files changed, 6 insertions(+), 4 deletions(-) diff --git a/lib/src/blackfish/cli/__main__.py b/lib/src/blackfish/cli/__main__.py index b7e7a435..d3268b6f 100644 --- a/lib/src/blackfish/cli/__main__.py +++ b/lib/src/blackfish/cli/__main__.py @@ -405,7 +405,7 @@ async def check_tigerflow_images() -> None: type=str, default=None, help=( - "Pin the container image, e.g. 'vllm/vllm-openai:v0.20.0'. Defaults to" + "Pin the container image, e.g. 'vllm/vllm-openai:v0.26.0'. Defaults to" " the configured image. See `blackfish ls` for the version in use." ), ) diff --git a/lib/src/blackfish/client.py b/lib/src/blackfish/client.py index f8de6e4a..d9461c24 100644 --- a/lib/src/blackfish/client.py +++ b/lib/src/blackfish/client.py @@ -305,7 +305,7 @@ async def async_launch_service( mount: Optional directory to mount grace_period: Time in seconds to wait before marking unhealthy image_ref: Pin the container image as "repo:tag" (e.g. - "vllm/vllm-openai:v0.20.0"). None uses the configured image, + "vllm/vllm-openai:v0.26.0"). None uses the configured image, which is then recorded on the service so restarts reuse it. auto_cleanup: If True, automatically stop and delete this service when the Python script exits (default: True) diff --git a/lib/src/blackfish/server/images.py b/lib/src/blackfish/server/images.py index dcc58c54..5e7f2d23 100644 --- a/lib/src/blackfish/server/images.py +++ b/lib/src/blackfish/server/images.py @@ -61,7 +61,7 @@ def resolve_image(image_ref: str | None, default: ImageSpec) -> ImageSpec: DEFAULT_IMAGES: dict[str, ImageSpec] = { - "text_generation": ImageSpec(repo="vllm/vllm-openai", tag="v0.20.0"), + "text_generation": ImageSpec(repo="vllm/vllm-openai", tag="v0.26.0"), "speech_recognition": ImageSpec( repo="ghcr.io/princeton-ddss/speech-recognition-inference", tag="0.2.1", diff --git a/lib/tests/cli/test_cli_run.py b/lib/tests/cli/test_cli_run.py index 3051e790..9ec0598a 100644 --- a/lib/tests/cli/test_cli_run.py +++ b/lib/tests/cli/test_cli_run.py @@ -1286,4 +1286,6 @@ def test_dry_run_renders_the_pinned_image( # The script itself, not the echo line above it. script = result.output.split("> image_ref:")[-1] assert "vllm-openai_v9.9.9" in script - assert "vllm-openai_v0.20.0" not in script + from blackfish.server.images import DEFAULT_IMAGES + + assert DEFAULT_IMAGES["text_generation"].sif not in script