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.
| Metrics | Qwen3.8-27B-Nvidia-Origin | Qwen3.8-27B-Mthreads-FlagOS |
|---|---|---|
| musr_murder_mysteries | 76.8 | 74.58 |
| GPQA_Diamond | 90.4 | 88.88 |
Environment Setup
| Item | Version |
|---|---|
| Docker Version | Docker version 27.5.1, build 9f9e405 |
| Operating System | 22.04.4 LTS (Jammy Jellyfish) |
docker pull harbor.baai.ac.cn/flagrelease-public/sglang-plugin-fl-qwen3.8-max-g7-mthreads001-gems5.0.2-treenone-cx0.13.0-pluginnone-vllmnone-cp310-pt29-musa43-x64-3.3.5-server:202608151039
pip install modelscope
modelscope download --model FlagRelease/Qwen3.8-27B-FP8-mthreads-FlagOS --local_dir /data/Qwen3.8-27B-FP8
docker run -itd --privileged --net host \
--name flagos \
-w /workspace \
-v /data/:/data/ \
-v /hsdatapool/zhiyuan/models/:/modesls/ \
-v /etc/localtime:/etc/localtime:ro \
-v /etc/timezone:/etc/timezone:ro \
--env MTHREADS_VISIBLE_DEVICES=all \
--shm-size=560g \
harbor.baai.ac.cn/flagrelease-public/sglang-plugin-fl-qwen3.8-max-g7-mthreads001-gems5.0.2-treenone-cx0.13.0-pluginnone-vllmnone-cp310-pt29-musa43-x64-3.3.5-server:202608151039 bash
export LD_LIBRARY_PATH=/usr/lib/x86_64-linux-gnu/:/usr/local/lib/:/usr/local/mtshmem/lib/:/usr/local/musa/lib:/usr/local/musa/lib:/usr/local/openmpi/lib:/usr/local/musa/mudnn/lib:
unset FLAGCX_PATH
export MCCL_IB_DISABLE=1
export MCCL_DEBUG=WARN
export SGLANG_FLAGGEMS_RECORD=1
export SGLANG_FLAGGEMS_LOG_PATH=/tmp/flaggems_op.txt
export SGLANG_FL_DISPATCH_DEBUG=1
export SGLANG_FL_DISPATCH_LOG=/tmp/dispatch.log
export SGLANG_FL_PREFER=flagos
export SGLANG_FL_FLAGOS_BLACKLIST=count_nonzero,index_put_,_index_put_impl_,index_select,mm,add,_unique2,unique_consecutive,bmm,index,floor_divide,topk,addmm,layer_norm,gelu,sigmoid,embedding
export SGLANG_FL_OOT_ENABLED=1
启动命令
MODEL_PATH=/data/Qwen3.8-27B-FP8
PY=/root/.virtualenvs/sglang-0.5.6/bin/python
LOG=/tmp/sglang_27b.log
exec $PY -m sglang.launch_server \
--model-path $MODEL_PATH --trust-remote-code \
--host 0.0.0.0 --port 30000 --served-model-name Qwen3.8-27B-FP8 \
--attention-backend triton --linear-attn-prefill-backend triton --linear-attn-decode-backend triton \
--sampling-backend flashinfer --disable-radix-cache --disable-piecewise-cuda-graph --mem-fraction-static 0.85 \
--schedule-conservativeness 0.0 --max-prefill-tokens 32768 --chunked-prefill-size 16384 \
--num-continuous-decode-steps 2 --triton-attention-num-kv-splits 16 \
--context-length 262144 --disable-cuda-graph --watchdog-timeout 1800 \
--model-loader-extra-config '{"enable_multithread_load": true, "num_threads": 128}' \
> server.log 2>&1 &
curl http://127.0.0.1:30000/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "Qwen3.8-27B",
"messages": [
{
"role": "user",
"content": "用一句话介绍北京"
}
],
"max_tokens": 256,
"temperature": 0.7,
"chat_template_kwargs": {
"enable_thinking": false
}
}'
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 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 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 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 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 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:
We warmly welcome global developers to join us:
The model weights are derived from Qwen/Qwen3.8-27B-FP8 and are open‑sourced under the Apache License 2.0: https://www.apache.org/licenses/LICENSE-2.0.txt
outside.safetensors 6.0 GBlayers-10.safetensors 384 MBlayers-12.safetensors 384 MBlayers-13.safetensors 384 MBlayers-14.safetensors 384 MBlayers-16.safetensors 384 MBlayers-17.safetensors 384 MBlayers-18.safetensors 384 MBlayers-20.safetensors 384 MBlayers-21.safetensors 384 MBlayers-22.safetensors 384 MBlayers-24.safetensors 384 MBlayers-25.safetensors 384 MBlayers-26.safetensors 384 MBlayers-28.safetensors 384 MBlayers-29.safetensors 384 MBlayers-30.safetensors 384 MBlayers-32.safetensors 384 MBlayers-33.safetensors 384 MBlayers-34.safetensors 384 MBlayers-36.safetensors 384 MBlayers-37.safetensors 384 MBlayers-38.safetensors 384 MBlayers-40.safetensors 384 MBlayers-41.safetensors 384 MBlayers-42.safetensors 384 MBlayers-44.safetensors 384 MBlayers-45.safetensors 384 MBlayers-46.safetensors 384 MBlayers-48.safetensors 384 MBlayers-49.safetensors 384 MBlayers-50.safetensors 384 MBlayers-52.safetensors 384 MBlayers-53.safetensors 384 MBlayers-54.safetensors 384 MBlayers-56.safetensors 384 MBlayers-57.safetensors 384 MBlayers-58.safetensors 384 MBlayers-60.safetensors 384 MBlayers-61.safetensors 384 MBlayers-62.safetensors 384 MBlayers-0.safetensors 384 MBlayers-1.safetensors 384 MBlayers-2.safetensors 384 MBlayers-4.safetensors 384 MBlayers-5.safetensors 384 MBlayers-6.safetensors 384 MBlayers-8.safetensors 384 MBlayers-9.safetensors 384 MBlayers-11.safetensors 372 MBlayers-15.safetensors 372 MBlayers-19.safetensors 372 MBlayers-23.safetensors 372 MBlayers-27.safetensors 372 MBlayers-31.safetensors 372 MBlayers-35.safetensors 372 MBlayers-39.safetensors 372 MBlayers-43.safetensors 372 MBlayers-47.safetensors 372 MBlayers-51.safetensors 372 MBlayers-55.safetensors 372 MBlayers-59.safetensors 372 MBlayers-63.safetensors 372 MBlayers-3.safetensors 372 MBlayers-7.safetensors 372 MBHave the files? Submit a torrent.
Submit a torrentHave the complete files and permission to share them? Replace the example folder path. The script checks the files, makes a torrent and submits its magnet and checksums. Keep your torrent client seeding. Get your community key from your settings, and read the script before running it.
curl -fsSL https://pirateface.co/package.sh | bash -s -- --repo FlagRelease/Qwen3.8-27B-FP8-mthreads-FlagOS ./model-folder