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

ROCm: bump 6.3.0 (#3259)

Browse Source
This commit is contained in:
HAI
2025-02-02 12:13:40 -08:00
committed by GitHub
parent 55f5fc68ac
commit 566d61d90f
7 changed files with 28 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
.github/workflows/release-docker-amd.yml vendored
View File

@@ -14,7 +14,7 @@ jobs:
environment: 'prod' environment: 'prod'
strategy: strategy:
matrix: matrix:
rocm_version: ['6.2.0'] rocm_version: ['6.3.0']
build_type: ['all', 'srt'] build_type: ['all', 'srt']
steps: steps:
- name: Checkout repository - name: Checkout repository
@@ -41,8 +41,8 @@ jobs:
run: | run: |
version=$(cat python/sglang/version.py | cut -d'"' -f2) version=$(cat python/sglang/version.py | cut -d'"' -f2)
if [ "${{ matrix.rocm_version }}" = "6.2.0" ]; then if [ "${{ matrix.rocm_version }}" = "6.3.0" ]; then
rocm_tag="rocm620" rocm_tag="rocm630"
else else
echo "Unsupported ROCm version" echo "Unsupported ROCm version"
exit 1 exit 1

4
docker/Dockerfile.rocm
View File

@@ -1,8 +1,8 @@
# Usage (to build SGLang ROCm docker image): # Usage (to build SGLang ROCm docker image):
# docker build --build-arg SGL_BRANCH=v0.4.2.post1 -t v0.4.2.post1-rocm620 -f Dockerfile.rocm . # docker build --build-arg SGL_BRANCH=v0.4.2.post1 -t v0.4.2.post1-rocm630 -f Dockerfile.rocm .
# default base image # default base image
ARG BASE_IMAGE="rocmshared/vllm-rocm:20250114-tuned-elementwise-layernorm" ARG BASE_IMAGE="rocm/vllm-dev:20250114"
FROM $BASE_IMAGE AS base FROM $BASE_IMAGE AS base
USER root USER root

4
docs/developer/setup_github_runner.md
View File

@@ -11,9 +11,9 @@ docker pull nvidia/cuda:12.1.1-devel-ubuntu22.04
# Nvidia # Nvidia
docker run --shm-size 128g -it -v /tmp/huggingface:/hf_home --gpus all nvidia/cuda:12.1.1-devel-ubuntu22.04 /bin/bash docker run --shm-size 128g -it -v /tmp/huggingface:/hf_home --gpus all nvidia/cuda:12.1.1-devel-ubuntu22.04 /bin/bash
# AMD # AMD
docker run --rm --device=/dev/kfd --device=/dev/dri --group-add video --shm-size 128g -it -v /tmp/huggingface:/hf_home lmsysorg/sglang:v0.4.2.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:v0.4.2.post1-rocm630 /bin/bash
# AMD just the last 2 GPUs # 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:v0.4.2.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:v0.4.2.post1-rocm630 /bin/bash
``` ```
### Step 2: Configure the runner by `config.sh` ### Step 2: Configure the runner by `config.sh`

6
docs/start/install.md
View File

@@ -54,7 +54,7 @@ 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: 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 ```bash
docker build --build-arg SGL_BRANCH=v0.4.2.post1 -t v0.4.2.post1-rocm620 -f Dockerfile.rocm . docker build --build-arg SGL_BRANCH=v0.4.2.post1 -t v0.4.2.post1-rocm630 -f Dockerfile.rocm .
alias drun='docker run -it --rm --network=host --device=/dev/kfd --device=/dev/dri --ipc=host \ 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 \ --shm-size 16G --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined \
@@ -63,11 +63,11 @@ alias drun='docker run -it --rm --network=host --device=/dev/kfd --device=/dev/d
drun -p 30000:30000 \ drun -p 30000:30000 \
-v ~/.cache/huggingface:/root/.cache/huggingface \ -v ~/.cache/huggingface:/root/.cache/huggingface \
--env "HF_TOKEN=<secret>" \ --env "HF_TOKEN=<secret>" \
v0.4.2.post1-rocm620 \ v0.4.2.post1-rocm630 \
python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --host 0.0.0.0 --port 30000 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 # Till flashinfer backend available, --attention-backend triton --sampling-backend pytorch are set by default
drun v0.4.2.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 v0.4.2.post1-rocm630 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 ## Method 4: Using docker compose

18
python/pyproject.toml
View File

@@ -19,31 +19,29 @@ dependencies = ["requests", "tqdm", "numpy", "IPython", "setproctitle"]
runtime_common = [ runtime_common = [
"aiohttp", "decord", "fastapi", "aiohttp", "decord", "fastapi",
"hf_transfer", "huggingface_hub", "interegular", "modelscope", "hf_transfer", "huggingface_hub", "interegular", "modelscope",
"orjson", "outlines>=0.0.44,<0.1.0", "orjson", "packaging", "pillow", "prometheus-client>=0.20.0",
"packaging", "pillow", "prometheus-client>=0.20.0", "psutil", "pydantic", "python-multipart", "pyzmq>=25.1.2",
"psutil", "pydantic", "python-multipart", "torchao>=0.7.0", "uvicorn", "uvloop", "xgrammar>=0.1.10"
"pyzmq>=25.1.2", "torchao>=0.7.0", "uvicorn", "uvloop",
"xgrammar>=0.1.10"
] ]
srt = [ srt = [
"sglang[runtime_common]", "cuda-python", "sglang[runtime_common]", "cuda-python",
"sgl-kernel>=0.0.3.post1", "torch", "vllm==0.6.4.post1", "sgl-kernel>=0.0.3.post1", "torch", "vllm==0.6.4.post1",
"flashinfer==0.1.6" "flashinfer==0.1.6", "outlines>=0.0.44,<0.1.0"
] ]
# HIP (Heterogeneous-computing Interface for Portability) for AMD # HIP (Heterogeneous-computing Interface for Portability) for AMD
# => base docker rocm/vllm-dev:20241022, not from public vllm whl # => base docker rocm/vllm-dev:20241022, not from public vllm whl
srt_hip = ["sglang[runtime_common]", "torch", "vllm==0.6.3.post2.dev1"] srt_hip = ["sglang[runtime_common]", "torch", "vllm==0.6.7.dev2", "outlines==0.1.11"]
# xpu is not enabled in public vllm and torch whl, # xpu is not enabled in public vllm and torch whl,
# need to follow https://docs.vllm.ai/en/latest/getting_started/xpu-installation.htmlinstall vllm # need to follow https://docs.vllm.ai/en/latest/getting_started/xpu-installation.htmlinstall vllm
srt_xpu = ["sglang[runtime_common]"] srt_xpu = ["sglang[runtime_common]", "outlines>=0.0.44,<0.1.0"]
#For Intel Gaudi(device : hpu) follow the installation guide #For Intel Gaudi(device : hpu) follow the installation guide
#https://docs.vllm.ai/en/latest/getting_started/gaudi-installation.html #https://docs.vllm.ai/en/latest/getting_started/gaudi-installation.html
srt_hpu = ["sglang[runtime_common]"] srt_hpu = ["sglang[runtime_common]", "outlines>=0.0.44,<0.1.0"]
# CPU: currently, there are no pre-built vllm wheels for CPU. # CPU: currently, there are no pre-built vllm wheels for CPU.
# To install vllm for CPU, please follow the instruction here: # To install vllm for CPU, please follow the instruction here:
# https://docs.vllm.ai/en/latest/getting_started/installation/cpu/index.html # https://docs.vllm.ai/en/latest/getting_started/installation/cpu/index.html
srt_cpu = ["sglang[runtime_common]", "torch"] srt_cpu = ["sglang[runtime_common]", "torch", "outlines>=0.0.44,<0.1.0"]
openai = ["openai>=1.0", "tiktoken"] openai = ["openai>=1.0", "tiktoken"]
anthropic = ["anthropic>=0.20.0"] anthropic = ["anthropic>=0.20.0"]

10
python/sglang/srt/constrained/outlines_backend.py
View File

@@ -20,7 +20,6 @@ from typing import Dict, List, Optional, Tuple, Union
import interegular import interegular
import torch import torch
from outlines.fsm.guide import RegexGuide from outlines.fsm.guide import RegexGuide
from outlines.fsm.json_schema import build_regex_from_schema
from outlines.models.transformers import TransformerTokenizer from outlines.models.transformers import TransformerTokenizer
from pydantic import BaseModel from pydantic import BaseModel
@@ -29,6 +28,15 @@ from sglang.srt.constrained.base_grammar_backend import (
BaseGrammarObject, BaseGrammarObject,
) )
from sglang.srt.constrained.outlines_jump_forward import OutlinesJumpForwardMap from sglang.srt.constrained.outlines_jump_forward import OutlinesJumpForwardMap
from sglang.srt.utils import is_hip
is_hip_ = is_hip()
if is_hip_:
from outlines_core.fsm.json_schema import build_regex_from_schema
else:
from outlines.fsm.json_schema import build_regex_from_schema
logger = logging.getLogger(__name__) logger = logging.getLogger(__name__)

2
python/sglang/srt/custom_op.py
View File

@@ -20,7 +20,7 @@ class CustomOp(nn.Module):
raise NotImplementedError raise NotImplementedError
def forward_hip(self, *args, **kwargs): def forward_hip(self, *args, **kwargs):
raise NotImplementedError return self.forward_native(*args, **kwargs)
def forward_xpu(self, *args, **kwargs): def forward_xpu(self, *args, **kwargs):
return self.forward_native(*args, **kwargs) return self.forward_native(*args, **kwargs)