Understand the project
Practical overview
When it fits
Use the repository when selecting an ERNIE 4.5 model or investigating a PaddlePaddle-based training workflow with ERNIEKit.
Who it is for
- Developers exploring language or vision-language models.
- Teams building PaddlePaddle-based model training or fine-tuning workflows.
Documented uses
- Review ERNIE 4.5 model variants and their input and output modalities.
- Find ERNIEKit resources for supervised fine-tuning and function-call training.
- Follow repository links to model, training, and deployment resources.
Before you choose
Requirements & limits
The repository and models are identified as open source under Apache 2.0.
- The listed language models take text as input and produce text; vision-language variants accept text, images, and video and produce text.
- The repository lists a 128K context window for the models shown.
- ERNIEKit is based on PaddlePaddle, and deployment resources are linked separately through PaddlePaddle FastDeploy.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Source-based practical guide
Train an ERNIE 4.5 0.3B model with ERNIEKit
Download the listed model and start supervised fine-tuning with the provided 8K sequence-length configuration.
Before you start
- ERNIEKit installed successfully.
- Access to download the Hugging Face model baidu/ERNIE-4.5-0.3B-Paddle.
Steps
Download baidu/ERNIE-4.5-0.3B-Paddle to the local directory baidu/ERNIE-4.5-0.3B-Paddle using huggingface-cli.
Start training with erniekit train examples/configs/ERNIE-4.5-0.3B/sft/run_sft_8k.yaml.
Expected outcome
ERNIEKit begins training the model using the provided SFT configuration.
Guide limitations
- For installation and further CLI guidance, consult the ERNIEKit Training Document.
Evidence and transparency
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 7 October 2026
- Original website title
- PaddlePaddle/ERNIE · Official homepage title · retained in its source language