Pangu Models by Huawei Noah's Ark Lab
About
The Pangu large language model (LLM) family, developed by Huawei, comprises advanced models tailored for various natural language processing tasks. Launched in 2021 as pioneering pre-trained Chinese LLMs with over 100 billion parameters 7, Pangu models have seen remarkable evolution. Pangu-α, featuring 200 billion parameters, was adeptly trained on 2048 Ascend processors through a hybrid parallel training technique 3. The series also includes Pangu-π, which enhances model performance with architectural innovations for increased nonlinearity and expressive power 2, and is utilized in specialized applications such as the YunShan model for finance and law 2. Notably, Pangu-Σ, with a trillion parameters, utilizes a sparse heterogeneous computing architecture 8, while Pangu 3.0 targets industry-specific needs with a comprehensive three-layer structure 5. This LLM family exemplifies Huawei's dedication to producing robust, efficient models for varied uses, though details of training and parameter specifications might differ across sources.
Current Variants
Use-when guidance is derived from seed capabilities, context, release, and replacement fields.
Use when the workload needs 128k context and 10B parameters.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Pangu 3.0 100B | Use when the workload needs 100B parameters. | 2024-04 | 100B parameters | Current |
| Pangu 3.0 71B | Use when the workload needs 71B parameters. | 2024-04 | 71B parameters | Current |
| Pangu 3.0 38B | Use when the workload needs 38B parameters. | 2024-04 | 38B parameters | Current |
| Pangu 3.0 10B | Use when the workload needs 128k context and 10B parameters. | 2024-04 | 128k context10B parameters | Current |
Release Timeline
1 release groupSpecifications(4 models)
| Model | Released | Context | Parameters |
|---|---|---|---|
| Pangu 3.0 100B | 2024-04 | — | 100B |
| Pangu 3.0 71B | 2024-04 | — | 71B |
| Pangu 3.0 38B | 2024-04 | — | 38B |
| Pangu 3.0 10B | 2024-04 | 128k | 10B |
Frequently Asked Questions
- What is Pangu used for?
- Pangu is used for coding. The family description and listed model capabilities point to those workloads as the best fit.
- How does Pangu compare to PixArt?
- Pangu by Huawei Noah's Ark Lab is strongest where you need coding, while PixArt by Huawei Noah's Ark Lab is the closest related family to check for image generation. Pangu has 4 listed variants and reaches up to 128k context, so compare the specs and pricing tables before choosing a production model.
- Which Pangu model should I use?
- If price is the main constraint, use the pricing table first because Pangu does not have complete provider pricing in the local data. For the most capable/latest local choice, evaluate Pangu 3.0 10B with 128k context.
