#6363rd of 70 in Open models
ernie
Open-weight ERNIE 4.5 multimodal MoE models with ERNIEKit fine-tuning tools
- Stars
- 7.7k
- License
- Apache-2.0
- Last commit
- Dec 2025
- Last release
- Jun 2025
- Language
- Python
7,653 on 20 Apr 20267,740 on 9 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
Baidu's ERNIE 4.5 repository covers ten open-weight models: MoE LLMs (300B-A47B, 21B-A3B), vision-language MoE models (424B-A47B, 28B-A3B) and a 0.3B dense model. It ships ERNIEKit, a PaddlePaddle toolkit for SFT, DPO, LoRA and QAT with CLI and WebUI. Inference is handled by the separate FastDeploy project, which provides an OpenAI-compatible server.
Who it is for: Teams fine-tuning or serving Baidu's ERNIE 4.5 models on PaddlePaddle
Strengths
- Apache-2.0 weights across 10 variants, from 0.3B dense to 424B MoE
- ERNIEKit covers SFT, DPO, LoRA variants and QAT with CLI and WebUI
- VL models support thinking and non-thinking modes; 128K context window
- Cookbook notebooks with Gradio demos for chat, search, RAG, OCR and fine-tuning
Weaknesses
- Training and inference depend on PaddlePaddle rather than PyTorch-native tooling
- Inference lives in a separate project, FastDeploy, so setup spans two repos
- Training for the ERNIE-4.5-VL base and 424B models was listed as coming soon
- Post-training quantization (PTQ) is marked work in progress; last release was June 2025
What it needs
- GPU required
- Docker
- Needs PaddlePaddle, FastDeploy, Hugging Face CLI
- Models: ERNIE-4.5-300B-A47B, ERNIE-4.5-21B-A3B, ERNIE-4.5-VL-424B-A47B, ERNIE-4.5-VL-28B-A3B, ERNIE-4.5-0.3B
- port 9904
What changed
From the release notes. No flag means the note does not say so, not that the upgrade is safe.
Quick answers
Is ernie open source?
Yes. Its code is on GitHub under the Apache-2.0 license.
Can you run ernie with Docker?
Yes. Its repository has a Docker setup: a Dockerfile, a Compose file or a published image.
Does ernie need a GPU?
Yes, it needs a GPU.
Also in Open models
See all 70| Rank | Project | Score |
|---|---|---|
| 1 | ruviewWiFi CSI sensing on ESP32 for presence, breathing, heart rate and pose | 94 out of 100 |
| 2 | layaSingle-pass multilingual classifier for typed choice, score and yes/no decisions | 88 out of 100 |
| 3 | rf-detrReal-time transformer models for object detection, segmentation and keypoints | 79 out of 100 |
| #4 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 78 out of 100 |
| #5 | yolov5PyTorch YOLOv5 object detection, segmentation and classification models with training code | 76 out of 100 |