Hugging Face's logo

openpangu
/
openPangu-2.0-Flash-Int8

Safetensors
openpangu_v2
custom_code
8-bit precision
compressed-tensors
Model card Files
xet
 Community
openPangu-2.0-Flash-Int8 / README_EN.md
Tian-Fantasea's picture
Tian-Fantasea
Update LICENSE Link
3d57106 verified
preview code
|
Raw
History Blame Contribute Delete
3.54 kB

openPangu-2.0-Flash-Int8

中文 | English

1. Introduction

openPangu-2.0-Flash is an MoE model trained on Ascend. The model has 92B total parameters and 6B activated parameters. Its context length is 512k. The total pretraining data contains 34T tokens. During Post-training, openPangu-2.0-Flash is trained through unified SFT with slow and fast thinking capability, multiple specialist RL traning, on-policy distillation combining multiple RL specialists. openPangu-2.0-Flash-Int8 is a quantized version of openPangu-2.0-Flash.

2. Architecture

openPangu-2.0-Flash-Int8 brings several major architectural improvements:

  • Efficient attention: The model retains MLA for efficient inference and combines DSA and SWA in a 1:2 layer ratio. SWA layers handle local-window modeling, while DSA layers capture sparse global context. This design lowers compute, memory footprint, and memory access costs for long-context inference while preserving accuracy.
  • Residual topology: The conventional residual path is replaced with a 4-stream mHC design, improving representation diversity and generalization.
  • Multi-token prediction (MTP): The model uses three MTP heads to draft 3 additional tokens per step, enabling faster inference through self-speculative decoding.
  • Optimizer: Training uses the Muon optimizer for faster convergence.

3. Deployment

4. Model License

Unless otherwise noted, the openPangu-2.0-Flash-Int8 model is licensed under the terms and conditions of OPENPANGU MODEL LICENSE AGREEMENT VERSION 2.0, which is intended to be used permissively and enable the further development of artificial intelligence technologies. Please refer to the LICENSE file located in the root directory of the model repository for details.

5. Disclaimer

Due to the technical limitations inherent in the technology on which the openPangu-2.0-Flash-Int8 model (“Model”) relies and the fact that the artificial intelligence generated content is automatically produced by Model, Huawei cannot make any guarantees regarding the following matters:

  • The output of this Model is automatically generated via AI algorithms, it does not rule out the possibility that some of the information may be flawed, unreasonable, or cause discomfort, and the generated content does not represent Huawei's attitude or standpoint;
  • There is no guarantee that this Model is 100% accurate, reliable, functional, timely, secure and safety, error-free, uninterrupted, continuously stable, or free of any faults;
  • The output of this Model does not constitute any advices or decisions for you, and it does not guarantee the authenticity, completeness, accuracy, timeliness, legality, functionality, or practicality of the generated content. The generated content cannot replace professionals in medical, legal, and other fields in answering your questions. The generated content is for your reference only and does not represent any attitude, standpoint, or position of Huawei. You need to make independent judgments based on your actual situation, and Huawei does not assume any responsibilities.

6. Contact

If you have any question, please raise an issue or contact us at openPangu@huawei.com.