openPangu 2.0 Pro
openPangu 2.0 Pro is Huawei's 505B-total, 18B-active mixture-of-experts language model with a 512K context window and published weights for Ascend-oriented deployment.
Platform: Replicate
Language ModelMixture of ExpertsLong ContextAscend
Huawei Ascend accelerator cluster
Open weights under the custom openPangu Model License Agreement 2.0; infrastructure costs vary
License Check Required🚀Function Overview
A cluster-scale open-weight model for long-context reasoning, agent tasks, and Ascend deployment experiments, distributed under the openPangu Model License Agreement 2.0.
Key Features
- 505B total parameters with 18B activated per token
- 512K-token context window
- Thinking and non-thinking response modes
- Official weights on GitCode and Hugging Face
- omni-infer deployment path for Huawei Ascend hardware
- Custom openPangu Model License Agreement 2.0 rather than Apache 2.0
Use Cases
- •Evaluating long-context reasoning and agent workflows
- •Testing large MoE inference on Huawei Ascend infrastructure
- •Comparing thinking and non-thinking modes on matched tasks
- •Studying a non-CUDA cluster-scale serving stack
⚙️Input Parameters
messages
arrayText chat messages formatted with the published openPangu tokenizer template.
thinking
booleanControls thinking or non-thinking generation in supported serving configurations.
tools
arrayOptional tool definitions supported by the published chat template and compatible serving stack.
💡Usage Examples
Example 1
Input Parameters
{
"messages": [
{
"role": "user",
"content": "Analyze this long technical record and list the unresolved assumptions."
}
],
"thinking": true
}Output Results
Generated text from an operator-managed openPangu 2.0 Pro deployment.
Quick Actions
Technical Specifications
- Hardware Type
- Huawei Ascend accelerator cluster
- Commercial Use
- Unknown/Restricted
- Pricing
- Open weights under the custom openPangu Model License Agreement 2.0; infrastructure costs vary
- Platform
- Replicate