ERNIE Models — Hardware Requirements
7 ERNIE models from Baidu and the community, from the smallest that runs in 0.7 GB of VRAM at Q4 up to 286.2B parameters. Every row links to full quantization tables and GPU compatibility.
All ERNIE Models by Size
| Model | Params | Runs from (Q4) | Context | Publisher | Quant downloads |
|---|---|---|---|---|---|
| ERNIE 4.5 0.3B PT | 361M | 0.7 GB | 131K | ||
| ERNIE 4.5 21B A3B PT | 21B | 13.8 GB | 131K | ||
| ERNIE 4.5 21B A3B Thinking | 21.8B | 14.3 GB | 131K | ||
| ERNIE 4.5 VL 28B A3B PT | 29.4B | 23.1 GB | 131K | ||
| ERNIE 5.5 Flash Base | 286.2B | 195.9 GB | 1049K |
Community variants and fine-tunes(2)
| Model | Params | Runs from (Q4) | Context | Publisher | Quant downloads |
|---|---|---|---|---|---|
| ERNIE 4.5 0.3B Paddle | 361M | 1.3 GB | 131K | ||
| ErniePEUnleashed | 3.4B | 3.7 GB | 262K |
Frequently Asked Questions
- How much VRAM do I need to run a ERNIE model?
- The smallest ERNIE model, ERNIE 4.5 0.3B PT, runs from 0.7 GB of VRAM at Q4_K_M (smaller low-bit quantizations start at 0.5 GB with a noticeable quality loss). Larger family members need proportionally more — see the table above for every model.
- Which ERNIE models can I run on a 16 GB GPU?
- 5 of 7 ERNIE models fit in 16 GB of VRAM at Q4_K_M, including ERNIE 4.5 21B A3B Thinking, ERNIE 4.5 21B A3B PT, ERNIE 4.5 0.3B PT.
- What is the most popular ERNIE model to run locally?
- ERNIE 4.5 21B A3B Thinking is the most downloaded ERNIE model in local-friendly quantized formats. It runs from 14.3 GB of VRAM at Q4_K_M.