Release v0.3.6.post2 (#2214)

Co-authored-by: Yineng Zhang <me@zhyncs.com>

docker/Dockerfile.rocm

 # Usage (to build SGLang ROCm docker image):
-#   docker build --build-arg SGL_BRANCH=v0.3.6.post1 -t testImage -f Dockerfile.rocm .
+#   docker build --build-arg SGL_BRANCH=v0.3.6.post2 -t testImage -f Dockerfile.rocm .
 
 # default base image
 ARG BASE_IMAGE="rocm/vllm-dev:20241022"

docs/developer/setup_github_runner.md

@@ -11,9 +11,9 @@ docker pull nvidia/cuda:12.1.1-devel-ubuntu22.04
 # Nvidia
 docker run --shm-size 128g -it -v /tmp/huggingface:/hf_home --gpus all nvidia/cuda:12.1.1-devel-ubuntu22.04 /bin/bash
 # AMD
-docker run --rm --device=/dev/kfd --device=/dev/dri --group-add video --shm-size 128g -it -v /tmp/huggingface:/hf_home lmsysorg/sglang:v0.3.6.post1-rocm620 /bin/bash
+docker run --rm --device=/dev/kfd --device=/dev/dri --group-add video --shm-size 128g -it -v /tmp/huggingface:/hf_home lmsysorg/sglang:v0.3.6.post2-rocm620 /bin/bash
 # AMD just the last 2 GPUs
-docker run --rm --device=/dev/kfd --device=/dev/dri/renderD176 --device=/dev/dri/renderD184 --group-add video --shm-size 128g -it -v /tmp/huggingface:/hf_home lmsysorg/sglang:v0.3.6.post1-rocm620 /bin/bash
+docker run --rm --device=/dev/kfd --device=/dev/dri/renderD176 --device=/dev/dri/renderD184 --group-add video --shm-size 128g -it -v /tmp/huggingface:/hf_home lmsysorg/sglang:v0.3.6.post2-rocm620 /bin/bash
 ```
 
 ### Step 2: Configure the runner by `config.sh`

docs/start/install.md

@@ -16,7 +16,7 @@ Note: Please check the [FlashInfer installation doc](https://docs.flashinfer.ai/
 ## Method 2: From source
 ```
 # Use the last release branch
-git clone -b v0.3.6.post1 https://github.com/sgl-project/sglang.git
+git clone -b v0.3.6.post2 https://github.com/sgl-project/sglang.git
 cd sglang
 
 pip install --upgrade pip
@@ -46,7 +46,7 @@ docker run --gpus all \
 Note: To AMD ROCm system with Instinct/MI GPUs, it is recommended to use `docker/Dockerfile.rocm` to build images, example and usage as below:
 
 ```bash
-docker build --build-arg SGL_BRANCH=v0.3.6.post1 -t v0.3.6.post1-rocm620 -f Dockerfile.rocm .
+docker build --build-arg SGL_BRANCH=v0.3.6.post2 -t v0.3.6.post2-rocm620 -f Dockerfile.rocm .
 
 alias drun='docker run -it --rm --network=host --device=/dev/kfd --device=/dev/dri --ipc=host \
     --shm-size 16G --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined \
@@ -55,11 +55,11 @@ alias drun='docker run -it --rm --network=host --device=/dev/kfd --device=/dev/d
 drun -p 30000:30000 \
     -v ~/.cache/huggingface:/root/.cache/huggingface \
     --env "HF_TOKEN=<secret>" \
-    v0.3.6.post1-rocm620 \
+    v0.3.6.post2-rocm620 \
     python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --host 0.0.0.0 --port 30000
 
 # Till flashinfer backend available, --attention-backend triton --sampling-backend pytorch are set by default
-drun v0.3.6.post1-rocm620 python3 -m sglang.bench_one_batch --batch-size 32 --input 1024 --output 128 --model amd/Meta-Llama-3.1-8B-Instruct-FP8-KV --tp 8 --quantization fp8
+drun v0.3.6.post2-rocm620 python3 -m sglang.bench_one_batch --batch-size 32 --input 1024 --output 128 --model amd/Meta-Llama-3.1-8B-Instruct-FP8-KV --tp 8 --quantization fp8
 ```
 
 ## Method 4: Using docker compose

python/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "sglang"
-version = "0.3.6.post1"
+version = "0.3.6.post2"
 description = "SGLang is yet another fast serving framework for large language models and vision language models."
 readme = "README.md"
 requires-python = ">=3.8"

python/sglang/bench_one_batch.py

@@ -466,7 +466,6 @@ if __name__ == "__main__":
 
     try:
         main(server_args, bench_args)
-    except Exception as e:
-        raise e
     finally:
-        kill_child_process()
+        if server_args.tp_size != 1:
+            kill_child_process()

python/sglang/launch_server.py

@@ -11,7 +11,5 @@ if __name__ == "__main__":
 
     try:
         launch_server(server_args)
-    except Exception as e:
-        raise e
     finally:
         kill_child_process()

python/sglang/srt/utils.py

@@ -517,6 +517,11 @@ def monkey_patch_vllm_p2p_access_check(gpu_id: int):
 
     setattr(tgt, "gpu_p2p_access_check", lambda *arg, **kwargs: True)
 
+    # Suppress the warnings from this delete function when using sglang.bench_one_batch
+    from vllm.distributed.device_communicators.custom_all_reduce import CustomAllreduce
+
+    setattr(CustomAllreduce, "__del__", lambda *args, **kwargs: None)
+
 
 vllm_all_gather_backup = None
 

python/sglang/version.py

-__version__ = "0.3.6.post1"
+__version__ = "0.3.6.post2"

test/srt/test_nightly_human_eval.py

@@ -3,7 +3,6 @@ import shutil
 import signal
 import subprocess
 import unittest
-from types import SimpleNamespace
 
 from test_nightly_gsm8k_eval import launch_server, parse_models
 
@@ -13,9 +12,7 @@ from sglang.test.test_utils import (
     DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_FP8_TP2,
     DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_TP1,
     DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_TP2,
-    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
     DEFAULT_URL_FOR_TEST,
-    popen_launch_server,
 )