Zum Inhalt springen
ModelÜberprüftExtern

DeepSeek V4 Pro Hygon FlagOS

ArtOtherVerleger@FlagReleaseCBewertung

Modellquelle

Art
Other
Dateigröße
36,5 GB
Aktualisiert
15. Sept. 2026
Verleger
@FlagRelease
Aufgaben
Textgenerierung

Quellenauszug

Modellartefakte

64 Artefakte

model0-mp64.safetensors

safetensors · 36,5 GB · SHA-256 8e458fe597b1…5387

model1-mp64.safetensors

safetensors · 36,5 GB · SHA-256 c5ddc91bc6ac…ab6c

model10-mp64.safetensors

safetensors · 36,5 GB · SHA-256 dc2fd484a234…0c9a

model11-mp64.safetensors

safetensors · 36,5 GB · SHA-256 fe180880b452…9e0c

model12-mp64.safetensors

safetensors · 36,5 GB · SHA-256 f8ce3f89814a…71f5

model13-mp64.safetensors

safetensors · 36,5 GB · SHA-256 044cafcde859…ba6b

model14-mp64.safetensors

safetensors · 36,5 GB · SHA-256 40fea30e11ab…491b

model15-mp64.safetensors

safetensors · 36,5 GB · SHA-256 b8f0c9ed17eb…dec4

model16-mp64.safetensors

safetensors · 36,5 GB · SHA-256 3af9331c6296…915e

model17-mp64.safetensors

safetensors · 36,5 GB · SHA-256 38a1f10e4216…e91c

model18-mp64.safetensors

safetensors · 36,5 GB · SHA-256 48f72e7b1cde…a594

model19-mp64.safetensors

safetensors · 36,5 GB · SHA-256 2f483cf490ed…0572

model2-mp64.safetensors

safetensors · 36,5 GB · SHA-256 1d7a5ac5e71a…875c

model20-mp64.safetensors

safetensors · 36,5 GB · SHA-256 4edd1f0f9aac…5e57

model21-mp64.safetensors

safetensors · 36,5 GB · SHA-256 57c88d9a18c8…3550

model22-mp64.safetensors

safetensors · 36,5 GB · SHA-256 0a041f982eb2…ecfb

model23-mp64.safetensors

safetensors · 36,5 GB · SHA-256 2676dfd9a22e…d65d

model24-mp64.safetensors

safetensors · 36,5 GB · SHA-256 04c9a9c7eac6…abca

model25-mp64.safetensors

safetensors · 36,5 GB · SHA-256 e9fe477f4e39…1e49

model26-mp64.safetensors

safetensors · 36,5 GB · SHA-256 0a8482d522bf…57b6

model27-mp64.safetensors

safetensors · 36,5 GB · SHA-256 553263ab2b41…e2c0

model28-mp64.safetensors

safetensors · 36,5 GB · SHA-256 bc710626d647…fbf9

model29-mp64.safetensors

safetensors · 36,5 GB · SHA-256 8c5ad7110527…1938

model3-mp64.safetensors

safetensors · 36,5 GB · SHA-256 4438e14af337…4296

model30-mp64.safetensors

safetensors · 36,5 GB · SHA-256 d73af6045982…0f0c

model31-mp64.safetensors

safetensors · 36,5 GB · SHA-256 505f56b995c4…7b0f

model32-mp64.safetensors

safetensors · 36,5 GB · SHA-256 2c5a260f52cb…2e8a

model33-mp64.safetensors

safetensors · 36,5 GB · SHA-256 97c36008391e…dd71

model34-mp64.safetensors

safetensors · 36,5 GB · SHA-256 b811bde46515…e053

model35-mp64.safetensors

safetensors · 36,5 GB · SHA-256 3b954b1d4e71…460a

model36-mp64.safetensors

safetensors · 36,5 GB · SHA-256 4ece99cd798e…f67e

model37-mp64.safetensors

safetensors · 36,5 GB · SHA-256 68be5ac21fcd…20f8

model38-mp64.safetensors

safetensors · 36,5 GB · SHA-256 74cbc6e73952…cfda

model39-mp64.safetensors

safetensors · 36,5 GB · SHA-256 f7f9c99b613f…d3ff

model4-mp64.safetensors

safetensors · 36,5 GB · SHA-256 09e958a27c86…c1df

model40-mp64.safetensors

safetensors · 36,5 GB · SHA-256 8eebfe667f3d…eb7a

model41-mp64.safetensors

safetensors · 36,5 GB · SHA-256 420f7d96cee5…e9db

model42-mp64.safetensors

safetensors · 36,5 GB · SHA-256 95011a760d7a…2910

model43-mp64.safetensors

safetensors · 36,5 GB · SHA-256 6eeb7cf3fcea…7fc0

model44-mp64.safetensors

safetensors · 36,5 GB · SHA-256 63d2c5a724f2…3d20

model45-mp64.safetensors

safetensors · 36,5 GB · SHA-256 4c72006fe38f…576d

model46-mp64.safetensors

safetensors · 36,5 GB · SHA-256 1c124c5d63ed…76b4

model47-mp64.safetensors

safetensors · 36,5 GB · SHA-256 8c7b3bc5de8f…6ab5

model48-mp64.safetensors

safetensors · 36,5 GB · SHA-256 16a60aa92eb4…0aeb

model49-mp64.safetensors

safetensors · 36,5 GB · SHA-256 09cc81a1cbb2…ba8d

model5-mp64.safetensors

safetensors · 36,5 GB · SHA-256 0b89b5382921…9887

model50-mp64.safetensors

safetensors · 36,5 GB · SHA-256 561ab948c8a7…2621

model51-mp64.safetensors

safetensors · 36,5 GB · SHA-256 8c7e1e469bff…fa96

model52-mp64.safetensors

safetensors · 36,5 GB · SHA-256 d2be95ca1db9…c330

model53-mp64.safetensors

safetensors · 36,5 GB · SHA-256 b3bef1e79f9a…faeb

model54-mp64.safetensors

safetensors · 36,5 GB · SHA-256 30d27e4c0eab…0a58

model55-mp64.safetensors

safetensors · 36,5 GB · SHA-256 f816a58bdb87…6908

model56-mp64.safetensors

safetensors · 36,5 GB · SHA-256 6f9611ea591b…d7e5

model57-mp64.safetensors

safetensors · 36,5 GB · SHA-256 1cfc6975b802…8097

model58-mp64.safetensors

safetensors · 36,5 GB · SHA-256 210c0bbbdd42…bfa9

model59-mp64.safetensors

safetensors · 36,5 GB · SHA-256 c0f688b43438…d8c7

model6-mp64.safetensors

safetensors · 36,5 GB · SHA-256 4be102a54274…89ce

model60-mp64.safetensors

safetensors · 36,5 GB · SHA-256 4f8cd3347f27…8b90

model61-mp64.safetensors

safetensors · 36,5 GB · SHA-256 3978082d80ed…74a0

model62-mp64.safetensors

safetensors · 36,5 GB · SHA-256 8ae605f91806…1ffc

model63-mp64.safetensors

safetensors · 36,5 GB · SHA-256 84ac3e46e344…0fc8

model7-mp64.safetensors

safetensors · 36,5 GB · SHA-256 3dba8cb98f80…a8e7

model8-mp64.safetensors

safetensors · 36,5 GB · SHA-256 b0b215bf0369…6743

model9-mp64.safetensors

safetensors · 36,5 GB · SHA-256 3412aae9c7d1…76a9

Quellenauszüge

3 Auszüge

--- # Introduction DeepSeek-V4-Pro is the flagship model of the DeepSeek-V4 series, built on a MoE architecture with 1.6T total parameters and 49B activated parameters, supporting up to one million tokens of context. Architecturally, it employs a hybrid attention mechanism (CSA + HCA) that, in million-token scenarios, requires only 27% of the inference FLOPs and 10% of the KV cache compared to V3.2. It also introduces Manifold-Constrained Hyperconnections (mHC) to improve cross-layer signal propagation stability, and uses the Muon optimizer to enhance training efficiency. The model was pre-trained on 32T+ high-quality tokens. Its most powerful inference mode, Pro-Max, achieves top-tier results on coding benchmarks and substantially closes the gap with leading closed-source models on reasoning and agent tasks — making it one of the best open-source models currently available. ### Integrated Deployment - Out-of-the-box inference scripts with pre-configured hardware and software parameters - Released **FlagOS-Hygon** 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 | Metrics | DeepSeek-V4-Pro-Nvidia-Origin | DeepSeek-V4-Pro-Hygon-FlagOS | |--------------|--------------------------------|--------------------------------------| | GPQA | - | - | | Aime | - | - | # User Guide Environment Setup | Item | Version | |------------------|----------------------| | Docker Version | Docker version 20.10.24, build 297e128 | | Operating System | Sugon OS 8.9 | ## Operation Steps ### Download FlagOS Image ```bash docker pull harbor.baai.ac.cn/flagrelease-public/flagrelease-hygon-deepseek-v4-pro:202604250050 ``` ### Download Open-source Model Weights ```bash pip install modelscope modelscope download --model FlagRelease/DeepSeek-V4-Pro-hygon-FlagOS --local_dir /data/DeepSeek-V4-Pro-hygon-FlagOS ``` ### Start the Container ```bash export NCCL_SOCKET_IFNAME=eno1 export NCCL_IB_DISABLE=1 export NCCL_DEBUG=INFO export USE_FLAGGEMS=1 export NCCL_DEBUG=WARN export GEMS_VENDOR=hygon docker run \ --name flagos \ --network=host \ --ipc=host \ --device=/dev/kfd \ --device=/dev/mkfd \ --device=/dev/dri \ -v /opt/hyhal:/opt/hyhal \ -v /data:/d...

Source context: 0 downloads · 0 likes · Repo FlagRelease/DeepSeek-V4-Pro-hygon-FlagOS

Source context: 0 downloads · 0 likes · Repo FlagRelease/DeepSeek-V4-Pro-hygon-FlagOS