ERNIE 4.5 0.3B
The smallest ERNIE 4.5 language model uses a dense network. It provides a compact starting point for local text processing. This entry covers the post-trained checkpoint.
Model design and access
Its released chat checkpoint has received post-training. The PT suffix identifies Transformer-style PyTorch weights; Paddle releases are separate formats. The model card supports English and Chinese and reports a 131,072-token window.ERNIE-4.5-0.3B-PT official model card
Practical use
Practical guidance. Use narrow tasks with clear outputs, such as short labels or structured extraction, to assess whether this small model is sufficient. A long configured window does not give it the reasoning capacity of the larger models. Measure quality before reducing hardware.ERNIE-4.5-0.3B-PT official model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | ERNIE-4.5-0.3B-PT |
| Context window | 131,072 tokens Official checkpoint context window. Input and generation share this limit. ERNIE-4.5-0.3B-PT official model card |
| Parameters | 0.36B dense parametersERNIE-4.5-0.3B-PT official model card |
| Architecture | 18 layers; 16 query heads and two key/value headsERNIE-4.5-0.3B-PT official model card |
| License | Apache 2.0ERNIE-4.5-0.3B-PT official model card |
Related models
- ERNIE 4.5 0.3B Base
Baidu ยท ERNIE 4.5
Dense 300-million-parameter base checkpoint in the ERNIE 4.5 language family.
Launch date not confirmedDownloadable