EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Release v0.3.6.post2 (#2214)

Browse Source 
Co-authored-by: Yineng Zhang <me@zhyncs.com>
This commit is contained in:
Lianmin Zheng
2024-11-27 03:35:30 -08:00
committed by GitHub
parent fb6e04a0c2
commit fed4c6946a
9 changed files with 16 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
docker/Dockerfile.rocm
View File

@@ -1,5 +1,5 @@
# 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"

4
docs/developer/setup_github_runner.md
View File

@@ -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`

8
docs/start/install.md
View File

@@ -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

2
python/pyproject.toml
View File

@@ -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"

5
python/sglang/bench_one_batch.py
View File

@@ -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()

2
python/sglang/launch_server.py
View File

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

5
python/sglang/srt/utils.py
View File

@@ -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

2
python/sglang/version.py
View File

@@ -1 +1 @@
__version__ = "0.3.6.post1"
__version__ = "0.3.6.post2"

3
test/srt/test_nightly_human_eval.py
View File

@@ -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,
)