Introduction

Alibaba open‑sources the vision‑language model Qwen3.8‑27B, and the Zhongzhi (众智) FlagOS community simultaneously completes Day 0 multi‑chip adaptation; based on the FlagOS unified open‑source technology stack, Qwen3.8‑27B has finished multi‑chip adaptation, precision alignment and deployment verification across 11 AI chips including T-Head（平头哥）、NVIDIA（英伟达）、Moore Threads（摩尔线程）、MetaX（沐曦）、Kunlunxin（昆仑芯）、Ascend（华为昇腾）、Hygon（海光）、Iluvatar CoreX（天数智芯）、Tsingmicro（清微智能）、Enflame（燧原科技）and Sunrise（曦望）, among which NVIDIA and Moore Threads support FP8 precision deployment while the rest run on BF16, and FlagOS has for the first time extended the adaptation of the latest Qwen model to ARM edge‑side platforms with a W4A8 low‑bit version, enabling developers to directly obtain corresponding out‑of‑the‑box solutions.

Integrated Deployment

• Out-of-the-box inference scripts with pre-configured hardware and software parameters
• Released FlagOS-Nvidia container image supporting deployment within minutes

Consistency Validation

• Rigorously evaluated through benchmark testing: Performance and results from the FlagOS software stack are compared against native stacks on multiple public.

Evaluation Results

Benchmark Result

User Guide

Environment Setup

Operation Steps

Download FlagOS Image

docker pull harbor.baai.ac.cn/flagrelease-public/qwen3.8-27b-nvidia004-gems5.3.3-tree0.5.0-cxnone-plugin0.3.0-vllm0.24.0-cp312-pt211-cu129-x64-580.126.20:202608141515

Download Open-source Model Weights

pip install modelscope
modelscope download --model FlagRelease/Qwen3.8-27B-BF16-nvidia-FlagOS --local_dir /data/Qwen3.8-27B

Start the Container

set -euo pipefail

IMAGE=${IMAGE:-harbor.baai.ac.cn/flagrelease-public/qwen3.8-27b-nvidia004-gems5.3.3-tree0.5.0-cxnone-plugin0.3.0-vllm0.24.0-cp312-pt211-cu129-x64-580.126.20:202608141515}
CONTAINER=${CONTAINER:-qwen38-27b-flagos-single-gpu}
MODEL_ROOT=${MODEL_ROOT:-/data}
GPU_DEVICE=${GPU_DEVICE:-0}

driver_libcuda=$(readlink -f "$(ldconfig -p | awk '/libcuda\.so\.1 \(libc6,x86-64\)/ {print $NF; exit}')")
driver_libnvml=$(readlink -f "$(ldconfig -p | awk '/libnvidia-ml\.so\.1 \(libc6,x86-64\)/ {print $NF; exit}')")
driver_libptxjit=$(readlink -f "$(ldconfig -p | awk '/libnvidia-ptxjitcompiler\.so\.1 \(libc6,x86-64\)/ {print $NF; exit}')")
test -f "${driver_libcuda}"
test -f "${driver_libnvml}"
test -f "${driver_libptxjit}"

device_args=(--device "/dev/nvidia${GPU_DEVICE}:/dev/nvidia0")
for dev in /dev/nvidiactl /dev/nvidia-uvm /dev/nvidia-uvm-tools \
  /dev/nvidia-nvlink /dev/nvidia-caps/*; do
  [[ -e "${dev}" ]] && device_args+=(--device "${dev}:${dev}")
done

docker run -d \
  --name "${CONTAINER}" \
  --network host \
  --ipc host \
  --pids-limit=-1 \
  --shm-size=64g \
  --ulimit memlock=-1:-1 \
  --ulimit stack=67108864:67108864 \
  --cap-add SYS_PTRACE \
  --security-opt seccomp=unconfined \
  --runtime runc \
  "${device_args[@]}" \
  -e "RELEASE_HOST_GPU_DEVICE=${GPU_DEVICE}" \
  -v "${driver_libcuda}:/driver/libcuda.so.1:ro" \
  -v "${driver_libcuda}:/driver/libcuda.so:ro" \
  -v "${driver_libnvml}:/driver/libnvidia-ml.so.1:ro" \
  -v "${driver_libnvml}:/driver/libnvidia-ml.so:ro" \
  -v "${driver_libptxjit}:/driver/libnvidia-ptxjitcompiler.so.1:ro" \
  -v "${driver_libptxjit}:/driver/libnvidia-ptxjitcompiler.so:ro" \
  -v "${MODEL_ROOT}:/models:ro" \
  -v /data:/data \
  "${IMAGE}" -lc 'exec sleep infinity'
docker exec -it flagos /bin/bash

Start the Server

set -euo pipefail

CONTAINER=${CONTAINER:-qwen38-27b-flagos-single-gpu}
PORT=${PORT:-8000}

docker exec -d \
  -e CUDA_VISIBLE_DEVICES=0 \
  -e TRANSFORMERS_OFFLINE=1 \
  -e HF_HUB_OFFLINE=1 \
  -e TOKENIZERS_PARALLELISM=false \
  -e VLLM_PLUGINS=fl \
  -e USE_FLAGGEMS=1 \
  -e VLLM_FL_PREFER=flagos \
  -e VLLM_FL_PREFER_ENABLED=true \
  -e 'VLLM_FL_PER_OP=attention_backend=vendor|flagos|reference' \
  -e VLLM_FL_FLAGOS_BLACKLIST=index_put_,index_put,_index_put_impl_,nonzero,copy_,to_copy,index,mm,mm_out \
  -e TRITON_LIBCUDA_PATH=/driver \
  -e FLAGGEMS_ENABLE_OPLIST_PATH=/tmp/flaggems_enable_oplist.txt \
  -e PORT="${PORT}" \
  "${CONTAINER}" bash -lc '
    exec vllm serve /data/Qwen3.8-27B \
      --tokenizer /data/Qwen3.8-27B \
      --served-model-name Qwen/Qwen3.8-27B \
      --trust-remote-code \
      --tensor-parallel-size 1 \
      --pipeline-parallel-size 1 \
      --data-parallel-size 1 \
      --distributed-executor-backend mp \
      --dtype bfloat16 \
      --max-model-len 102400 \
      --max-num-seqs 64 \
      --max-num-batched-tokens 16384 \
      --gpu-memory-utilization 0.85 \
      --limit-mm-per-prompt "{\"image\":16}" \
      --seed 1234 \
      --no-enable-log-requests \
      --host 0.0.0.0 \
      --port "${PORT}" \
      >/tmp/qwen38-27b-flagos.log 2>&1
  '

for _ in $(seq 1 180); do
  if curl --noproxy '*' -fsS "http://127.0.0.1:${PORT}/health" >/dev/null; then
    echo "health=200 port=${PORT}"
    exit 0
  fi
  sleep 10
done

docker exec "${CONTAINER}" tail -n 200 /tmp/qwen38-27b-flagos.log >&2 || true
exit 1

Service Invocation

Invocation Script

set -euo pipefail

CONTAINER=${CONTAINER:-qwen38-27b-flagos-single-gpu}
PORT=${PORT:-8000}

docker exec -d \
  -e CUDA_VISIBLE_DEVICES=0 \
  -e TRANSFORMERS_OFFLINE=1 \
  -e HF_HUB_OFFLINE=1 \
  -e TOKENIZERS_PARALLELISM=false \
  -e VLLM_PLUGINS=fl \
  -e USE_FLAGGEMS=1 \
  -e VLLM_FL_PREFER=flagos \
  -e VLLM_FL_PREFER_ENABLED=true \
  -e 'VLLM_FL_PER_OP=attention_backend=vendor|flagos|reference' \
  -e VLLM_FL_FLAGOS_BLACKLIST=index_put_,index_put,_index_put_impl_,nonzero,copy_,to_copy,index,mm,mm_out \
  -e TRITON_LIBCUDA_PATH=/driver \
  -e FLAGGEMS_ENABLE_OPLIST_PATH=/tmp/flaggems_enable_oplist.txt \
  -e PORT="${PORT}" \
  "${CONTAINER}" bash -lc '
    exec vllm serve /models/Qwen3.8-27B \
      --tokenizer /models/Qwen3.8-27B \
      --served-model-name Qwen/Qwen3.8-27B \
      --trust-remote-code \
      --tensor-parallel-size 1 \
      --pipeline-parallel-size 1 \
      --data-parallel-size 1 \
      --distributed-executor-backend mp \
      --dtype bfloat16 \
      --max-model-len 102400 \
      --max-num-seqs 64 \
      --max-num-batched-tokens 16384 \
      --gpu-memory-utilization 0.85 \
      --limit-mm-per-prompt "{\"image\":16}" \
      --seed 1234 \
      --no-enable-log-requests \
      --host 0.0.0.0 \
      --port "${PORT}" \
      >/tmp/qwen38-27b-flagos.log 2>&1
  '

for _ in $(seq 1 180); do
  if curl --noproxy '*' -fsS "http://127.0.0.1:${PORT}/health" >/dev/null; then
    echo "health=200 port=${PORT}"
    exit 0
  fi
  sleep 10
done

docker exec "${CONTAINER}" tail -n 200 /tmp/qwen38-27b-flagos.log >&2 || true
exit 1

AnythingLLM Integration Guide

1. Download & Install

• Visit the official site: https://anythingllm.com/
• Choose the appropriate version for your OS (Windows/macOS/Linux)
• Follow the installation wizard to complete the setup

2. Configuration

• Launch AnythingLLM
• Open settings (bottom left, fourth tab)
• Configure core LLM parameters
• Click "Save Settings" to apply changes

3. Model Interaction

• After model loading is complete:
• Click "New Conversation"
• Enter your question (e.g., “Explain the basics of quantum computing”)
• Click the send button to get a response

Technical Overview

FlagOS is a fully open-source system software stack designed to unify the "model–system–chip" layers and foster an open, collaborative ecosystem. It enables a “develop once, run anywhere” workflow across diverse AI accelerators, unlocking hardware performance, eliminating fragmentation among vendor-specific software stacks, and substantially lowering the cost of porting and maintaining AI workloads. With core technologies such as the FlagScale, together with vllm-plugin-fl, distributed training/inference framework, FlagGems universal operator library, FlagCX communication library, and FlagTree unified compiler, the FlagRelease platform leverages the FlagOS stack to automatically produce and release various combinations of <chip + open-source model>. This enables efficient and automated model migration across diverse chips, opening a new chapter for large model deployment and application.

FlagGems

FlagGems is a high-performance, generic operator libraryimplemented in Triton language. It is built on a collection of backend-neutralkernels that aims to accelerate LLM (Large-Language Models) training and inference across diverse hardware platforms.

FlagTree

FlagTree is an open source, unified compiler for multipleAI chips project dedicated to developing a diverse ecosystem of AI chip compilers and related tooling platforms, thereby fostering and strengthening the upstream and downstream Triton ecosystem. Currently in its initial phase, the project aims to maintain compatibility with existing adaptation solutions while unifying the codebase to rapidly implement single-repository multi-backend support. Forupstream model users, it provides unified compilation capabilities across multiple backends; for downstream chip manufacturers, it offers examples of Triton ecosystem integration.

FlagScale and vllm-plugin-fl

Flagscale is a comprehensive toolkit designed to supportthe entire lifecycle of large models. It builds on the strengths of several prominent open-source projects, including Megatron-LM and vLLM, to provide a robust, end-to-end solution for managing and scaling large models. vllm-plugin-fl is a vLLM plugin built on the FlagOS unified multi-chip backend, to help flagscale support multi-chip on vllm framework.

FlagCX

FlagCX is a scalable and adaptive cross-chip communication library. It serves as a platform where developers, researchers, and AI engineers can collaborate on various projects, contribute to the development of cutting-edge AI solutions, and share their work with the global community.

FlagEval Evaluation Framework

FlagEval is a comprehensive evaluation system and open platform for large models launched in 2023. It aims to establish scientific, fair, and open benchmarks, methodologies, and tools to help researchers assess model and training algorithm performance. It features:

• Multi-dimensional Evaluation: Supports 800+ modelevaluations across NLP, CV, Audio, and Multimodal fields,covering 20+ downstream tasks including language understanding and image-text generation.
• Industry-Grade Use Cases: Has completed horizonta1 evaluations of mainstream large models, providing authoritative benchmarks for chip-model performance validation.

Contributing

We warmly welcome global developers to join us:

1. Submit Issues to report problems
2. Create Pull Requests to contribute code
3. Improve technical documentation
4. Expand hardware adaptation support

License

The model weights are derived from Qwen/Qwen3.8-27B and are open‑sourced under the Apache License 2.0: https://www.apache.org/licenses/LICENSE-2.0.txt