ERNIE 4.5 300B A47B Base
The large ERNIE 4.5 text model uses a sparse expert network for general language tasks. This entry covers the base checkpoint.
Model design and access
This is the separately published pretrained text checkpoint. It uses the same 300B total and 47B active scale as its chat counterpart, with a 131,072-token context. It is intended for further training rather than direct assistant behavior.ERNIE-4.5-300B-A47B-Base-PT official model card
Practical use
Practical guidance. Use the base only when control over training justifies the infrastructure. Record which data and alignment steps produce each derivative. That record makes it possible to distinguish a model limitation from a regression introduced by adaptation.ERNIE-4.5-300B-A47B-Base-PT official model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | ERNIE-4.5-300B-A47B-Base-PT |
| Context window | 131,072 tokens Official checkpoint context window. Input and generation share this limit. ERNIE-4.5-300B-A47B-Base-PT official model card |
| Parameters | 300B total parameters; 47B active per tokenERNIE-4.5-300B-A47B-Base-PT official model card |
| Architecture | 54 layers; 64 query heads and eight key/value headsERNIE-4.5-300B-A47B-Base-PT official model card |
| License | Apache 2.0ERNIE-4.5-300B-A47B-Base-PT official model card |
Related models
- ERNIE 4.5 300B A47B
Baidu ยท ERNIE 4.5
Post-trained sparse text model with 300 billion total parameters and 47 billion active parameters.
Launch date not confirmedDownloadable