|
--- |
|
license: apache-2.0 |
|
--- |
|
# β¨ Orion-7B |
|
**Orion is based on alpaca-7B, further trained on the latest deep learning papers published at notable conferences with various PEFT methods. It is made available under the Apache 2.0 license.** |
|
*Paper coming soon π.* |
|
|
|
|
|
## What is so special about Orion? |
|
* **Comprehensive performance analysis of PEFT methods for knowledge editing** |
|
* **Novel framekwork for effective knowledge editing dataset generation** |
|
* **Large knowledge editing dataset from recent deep learning papers published in top conferences** |
|
* |
|
β οΈ **This is a raw, fine-tuned model, which should be further instruction-tuned for production level performance.** |
|
|
|
π₯ **Orion LLM requires PyTorch 2.0** |
|
|
|
You will need **at least 20-25GB of memory** to swiftly run inference with Orion-7B. |
|
|
|
## Model Details |
|
|
|
### Model Description |
|
- **Developed by:** [AttentionX](https://attentionx.github.io); |
|
- **Base model:** alpaca-7B; |
|
- **Training method: LoRA, llama-adapter, llama-adapterV2, full fine-tuning** |
|
- **Language(s) (NLP):** English; |
|
- **License:** Apache 2.0 license. |
|
|
|
### Model Source |
|
- **Paper:** *coming soon*. |
|
|
|
|
|
## Training Details |
|
*Paper coming soon π.* |
|
|
|
|
|
## Evaluation |
|
*Paper coming soon.* |
|
### Test Inference |
|
1. What is GPT-4? |
|
1. llama |
|
- GPT-4 is a neural network model built by OpenAI, a non-profit artificial intelligence research organization founded in 2015 by Elon Musk, Sam Altman and other industry leaders. |
|
2. Orion |
|
- GPT-4 is a large-scale, multimodal model that can accept image and text inputs and produce text outputs. It is a language model developed by OpenAI. The GPT-4 base model has a larger model size and can accept image and text inputs |
|
2. What is the lowest score GPT-4 achieved in AP Calculus BC? |
|
1. llama |
|
- The lowest score achieved is 60% |
|
2. Orion |
|
- The lowest score achieved in AP Calculus BC (MCQ) is 33.33% |
|
|
|
|
|
## Hardware |
|
Orion was trained on a single a100 |
|
|
|
|
|
## License |
|
|
|
Orion is made available under the Apache 2.0 license. |
|
|
|
## Contact |
|
attentionx.ai@gmail.com |