Introduction

Tencent Hunyuan has officially open-sourced its multimodal model Hy4 Flash. Based on the unified multi-chip software stack FlagOS, the Zhongzhi FlagOS community has completed Day 0 cross-chip adaptation, covering nine chip vendors: T-Head, NVIDIA, Huawei Ascend, Hygon, Kunlunxin, Muxi, Moore Threads, TsingMicro and Iluvatar CoreX. Benefiting from the performance optimizations in FlagOS 2.2, overall evaluation results show that the FlagOS build running on NVIDIA H100 delivers up to a 35.9% increase in output throughput and up to a 29.1% reduction in time-to-first-token latency compared with the native NVIDIA stack. Industry users may deploy the FlagOS image of Hy4 Flash across diverse hardware by matching it with their existing computing resources and service workloads.

Hy4 Flash is designed for workloads including deep search, document processing, PPT generation and Computer Use for daily office scenarios. It has approximately 220 billion total parameters (including around 64 billion Engram parameters), with roughly 12 billion parameters activated per token. It natively supports text and image inputs and text outputs, alongside a maximum context length of 1 million tokens. Built with hybrid attention, Latent MoE and Engram architectures, its KV Cache footprint is only one-sixth that of Hy4. In Tencent Hunyuan’s internal WorkIndex benchmark, Hy4 Flash scores 42.1, outperforming Hy4 Preview’s 39.7, while boasting lower per-task costs.

Integrated Deployment

  • Out-of-the-box inference scripts with pre-configured hardware and software parameters
  • Released FlagOS-Mthreads 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 benchmarks.

Evaluation Results

Benchmark Result

Metrics Hy4-flash-Nvidia-Origin Hy4-flash-Mthreads-FlagOS
Math-500 96 Evaluating
GPQA_Diamond 92,42 Evaluating
mmbench 92.6 Evaluating
mmstar 80.93 Evaluating

User Guide

Environment Setup

Host Environment

Item Version / Spec
Hardware (GPU) MTT S5000 x8卡
Docker Version 24.0.9
Operating System 22.04.4 LTS
Driver 3.3.5-server
SDK 3.3.5-server

Operation Steps

Download FlagOS Image

docker pull harbor.baai.ac.cn/flagrelease-public/hy4-flash-mthreads001-gems5.3.0-tree0.6.1-cx0.13.0-pluginnone-vllm0.10.1-sglang0.0.0-sglangfl0.1.0-cp310-pt29-musa43-x64-3.3.5-server:202611190947

Download Open-source Model Weights

pip install modelscope
modelscope download --model FlagRelease/Hy4-flash-INT8-mthreads-FlagOS

Start the Container

docker run -dit \
  --name hy4flash-w8a8-tp8-release \
  --runtime mthreads \
  --privileged \
  --network host \
  --ipc host \
  --shm-size 1g \
  --ulimit memlock=-1:-1 \
  --ulimit stack=67108864:67108864 \
  --mount type=bind,src=/bjzyrgznyjy_ssd/models/Hy4-flash-testing-W8A8-no-MTP,dst=/models/Hy4-flash-testing-W8A8-no-MTP,readonly \
  -v /var/log/hy4flash-release:/var/log/hy4flash-release \
  -v /var/cache/hy4flash-release:/var/cache/hy4flash-release \
  --entrypoint /bin/sleep \
  harbor.baai.ac.cn/flagrelease-public/hy4-flash-mthreads001-gems5.3.0-tree0.6.1-cx0.13.0-pluginnone-vllm0.10.1-sglang0.0.0-sglangfl0.1.0-cp310-pt29-musa43-x64-3.3.5-server:202611190947 infinity
docker exec -it hy4flash-w8a8-tp8-release /bin/bash

Start the Inference Server

source /opt/hy4-preview-release/env.sh
export PYTHONPATH="/opt/hy4flash-release/runtime${PYTHONPATH:+:$PYTHONPATH}"
export USE_FLAGGEMS=1
export SGLANG_FL_FLAGOS_BLACKLIST='count_nonzero,index_put_,index_put_impl,index_select,mm,add,add_,mul,mul_,sum,sum_out,sum_dim,sum_dim_out,mean,mean_out,mean_dim,mean_dim_out,rsqrt,sigmoid,unique,_unique2,unique_consecutive,bmm,floor_divide,index,topk,copy_,_to_copy,to_copy,sort'
export SGLANG_FLAGGEMS_RECORD=1 SGLANG_FL_DISPATCH_DEBUG=1
export SGLANG_FLAGGEMS_LOG_PATH=/var/log/hy4flash-release/flaggems.log
export SGLANG_FL_DISPATCH_LOG=/var/log/hy4flash-release/dispatch.log
export TRITON_CACHE_DIR=/var/cache/hy4flash-release
export HY4_FLASH_FP32_REDUCE=1 SGLANG_MUSA_NATIVE_FLAGCX=1 FLAGCX_PATH=/sgl-workspace/FlagCX
export PYTORCH_MUSA_ALLOC_CONF=expandable_segments:False SGLANG_OPT_USE_TOPK_V2=0
export HY4_FLASH_VERIFY_GRAPH=0 HY4_FLASH_VERIFY_PREFILL_GRAPH=0 HY4_FLASH_GRAPH_TRACE=0
export HY4_FLASH_PACKED_HC=0 HY4_FLASH_GEMS_FWHT=1 HY4_FLASH_PREFILL_GRAPH=1 HY4_FLASH_GEMS_NORM=0 HY4_FLASH_MATE_GATE=0
export HY4_FLASH_NATIVE_SPARSE_HEADS=0 SGLANG_LOG_DECODE_GRAPH_KEY=0
export HY4_FLASH_EVIDENCE_BY_DEVICE=0 HY4_EVIDENCE_BY_DEVICE=0
export HY4_FLASH_TOKEN_KV_POOL=0 HY4_FLASH_LINEAR_NGRAM=0
export HY4_FLASH_FUSED_INDEX_EXPAND=1 HY4_FLASH_FUSED_INDEX_LOOKUP=1 HY4_FLASH_SHARED_SWA_INDICES=1 HY4_FLASH_PACKED_QKV=1
export HY4_FLASH_FUSED_ENGRAM_HISTORY=1 HY4_FLASH_WORD_RMS=1 HY4_FLASH_JOIN_NORM_POST=1 HY4_FLASH_MOE_KMAJOR=1
export HY4_FLASH_DENSE_KN=1 HY4_FLASH_WORD_HC_POST=1 HY4_FLASH_WORD_HC_POST_AUDIT=0
export HY4_FLASH_PREFILL_TP_HEADS=1 HY4_FLASH_TP_HEAD_REAL_ORACLE=''
export HY4_FLASH_PREFILL_ROUTER162=full-triton HY4_FLASH_PREFILL_TILE162=1 HY4_FLASH_PREFILL_162_AUDIT='' HY4_FLASH_PREFILL_ROUTE_RECEIPTS=''
export SGLANG_MOE_CONFIG_DIR=/opt/hy4flash-release/assets/moe-configs
export HY4_FLASH_LINEAR_CONFIG=/opt/hy4flash-release/assets/linear-config.json
export HY4_FLASH_DENSE_KN_PLAN=/opt/hy4flash-release/assets/dense-kn-plan.json
nohup /root/.virtualenvs/sglang-0.5.6/bin/python -m sglang.launch_server \
  --model-path /models/Hy4-flash-testing-W8A8-no-MTP --served-model-name Hy4-flash \
  --trust-remote-code --device musa --tp-size 8 --quantization w8a8_int8 --dtype bfloat16 \
  --host 0.0.0.0 --port 30200 --context-length 8192 --mem-fraction-static 0.85 \
  --attention-backend dsa --dsa-prefill-backend flashmla_sparse --dsa-decode-backend flashmla_sparse \
  --disable-custom-all-reduce --max-total-tokens 393216 --max-running-requests 64 \
  --cuda-graph-bs 1 2 4 8 16 32 64 --chunked-prefill-size 4096 \
  --disable-overlap-schedule --num-continuous-decode-steps 1 \
  --cuda-graph-config '{"decode":{"backend":"full","bs":[1,2,4,8,16,32,64]},"prefill":{"backend":"full","bs":[4096],"full_prefill_max_req":1}}' \
  --disable-chunked-prefix-cache \
  </dev/null >/var/log/hy4flash-release/server.log 2>&1 &

Service Invocation

Invocation Script (Text)

curl --fail-with-body --silent --show-error --max-time 180 \
  http://127.0.0.1:30200/v1/chat/completions \
  -H 'Content-Type: application/json' \
  --data-binary '{"model":"Hy4-flash","messages":[{"role":"user","content":"只回答数字:1+1等于多少?"}],"temperature":0,"max_tokens":32,"stream":false,"chat_template_kwargs":{"reasoning_effort":"no_think"}}'

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 Hy4-flash and are open-sourced under the Apache License 2.0: https://www.apache.org/licenses/LICENSE-2.0.txt

Downloads last month
-
Safetensors
Model size
220B params
Tensor type
I8
·
BF16
·
F32
·
I32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support