Skip to content

chore(sglang): bump to 0.5.12.post1 #9677

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ishandhanani merged 4 commits into from
May 25, 2026
Merged

chore(sglang): bump to 0.5.12.post1 #9677

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
d0a4669
sglang: bump to 0.5.12
ishandhanani May 18, 2026
62521d8
sglang: pin accelerate in [sglang] extra
ishandhanani May 18, 2026
ff4d59d
Merge main into idhanani/sgl-to-0.5.12 to pick up sglang 0.5.11 → 0.5…
dmitry-tokarev-nv May 21, 2026
1a222b3
chore(sglang): bump pin to 0.5.12.post1 for NIXL disagg fix
ishandhanani May 25, 2026
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
8 changes: 2 additions & 6 deletions components/src/dynamo/frontend/tests/test_sglang_processor_unit.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1298,10 +1298,6 @@ def test_system_message(self, tokenizer):
)
assert len(with_system.prompt_token_ids) > len(without_system.prompt_token_ids)

@pytest.mark.skip(
reason="DYN-3049: deepseek_v4 dispatch path requires sglang 0.5.12 support; "
"Dynamo is pinned to sglang 0.5.11. Unskip after the 0.5.12 bump lands."
)
def test_deepseek_v4_uses_sglang_encoder_when_chat_template_missing(
self, monkeypatch
):
Expand Down Expand Up @@ -1358,7 +1354,7 @@ def encode(self, prompt):
request,
tokenizer=NoTemplateTokenizer(),
tool_call_parser_name=None,
reasoning_parser_name="deepseek_v4",
reasoning_parser_name="deepseek-v4",
)

assert result.prompt_token_ids == [1, 2, 3]
Expand Down Expand Up @@ -1412,7 +1408,7 @@ def encode(self, prompt):
request,
tokenizer=NoTemplateTokenizer(),
tool_call_parser_name=None,
reasoning_parser_name="deepseek_v4",
reasoning_parser_name="deepseek-v4",
)

tools = captured["messages"][0]["tools"]
Expand Down
4 changes: 2 additions & 2 deletions container/compliance/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -128,8 +128,8 @@ python container/compliance/process_results.py \
|-----------|------|------------|
| `vllm` | 12.9 | `vllm/vllm-openai:v0.21.0-cu129-ubuntu2404` |
| `vllm` | 13.0 | `vllm/vllm-openai:v0.21.0-ubuntu2404` |
| `sglang` | 12.9 | `lmsysorg/sglang:v0.5.11-cu129-runtime` |
| `sglang` | 13.0 | `lmsysorg/sglang:v0.5.11-cu130-runtime` |
| `sglang` | 12.9 | `lmsysorg/sglang:v0.5.12.post1-cu129-runtime` |
| `sglang` | 13.0 | `lmsysorg/sglang:v0.5.12.post1-cu130-runtime` |
| `trtllm` | 13.1 | `nvcr.io/nvidia/cuda-dl-base:25.12-cuda13.1-runtime-ubuntu24.04` |
| `dynamo` frontend | — | `nvcr.io/nvidia/base/ubuntu:noble-20250619` |

Expand Down
4 changes: 2 additions & 2 deletions container/context.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -76,12 +76,12 @@ sglang:
base_image: nvcr.io/nvidia/cuda-dl-base
runtime_image: lmsysorg/sglang
base_image_tag: 25.06-cuda12.9-devel-ubuntu24.04
runtime_image_tag: v0.5.11-cu129-runtime
runtime_image_tag: v0.5.12.post1-cu129-runtime
cuda13.0:
base_image: nvcr.io/nvidia/cuda-dl-base
runtime_image: lmsysorg/sglang
base_image_tag: 25.11-cuda13.0-devel-ubuntu24.04
runtime_image_tag: v0.5.11-cu130-runtime
runtime_image_tag: v0.5.12.post1-cu130-runtime
# Builds the NIXL C++ SDK in wheel_builder so the dev stage can link nixl-sys
# (see templates/dev.Dockerfile). Runtime stays on the upstream lmsysorg/sglang
# NIXL Python stack — its wheel COPY is narrowed to ai_dynamo*.whl so the SDK
Expand Down
2 changes: 1 addition & 1 deletion container/templates/sglang_runtime.Dockerfile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,7 +59,7 @@ RUN --mount=type=cache,target=/root/.cache/pip,sharing=locked \
pip install --break-system-packages --no-deps "accelerate==1.13.0"

# Install distro: openai>=1.x's _base_client imports it unconditionally, and
# sglang 0.5.11's server_args eagerly imports sglang.srt.entrypoints.openai.protocol
# sglang 0.5.12's server_args eagerly imports sglang.srt.entrypoints.openai.protocol
# which pulls in openai.types.responses → triggers openai pkg init → import distro.
# The upstream lmsysorg/sglang runtime installs openai with --no-deps so distro is
# missing; without this any dynamo.sglang worker fails to import at startup.
Expand Down
4 changes: 3 additions & 1 deletion pyproject.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -63,7 +63,9 @@ vllm = [

sglang = [
"uvloop",
"sglang[diffusion]==0.5.11",
"sglang[diffusion]==0.5.12.post1",
# sglang[diffusion] dropped accelerate in 0.5.12; diffusers still needs it.
"accelerate>=0.17.0",
"nixl[cu12]>=1.0.0,<1.1.0",
"cupy-cuda12x>=13.0.0",
]
Expand Down
Loading