|
--- |
|
license: other |
|
license_name: stem.ai.mtl |
|
license_link: LICENSE |
|
language: |
|
- en |
|
tags: |
|
- phi-2 |
|
- electrical engineering |
|
- Microsoft |
|
datasets: |
|
- STEM-AI-mtl/Electrical-engineering |
|
- garage-bAInd/Open-Platypus |
|
task_categories: |
|
- question-answering |
|
- text-generation |
|
pipeline_tag: text-generation |
|
widget: |
|
- text: "Enter your instruction here" |
|
inference: true |
|
auto_sample: true |
|
inference_code: chat-GPTQ.py |
|
library_tag: transformers |
|
--- |
|
# For the electrical engineering community |
|
|
|
A unique, deployable and efficient 2.7 billion parameters model in the field of electrical engineering. This repo contains the adapters from the LoRa fine-tuning of the phi-2 model from Microsoft. It was trained on the [STEM-AI-mtl/Electrical-engineering](https://huggingface.co/datasets/STEM-AI-mtl/Electrical-engineering) dataset combined with [garage-bAInd/Open-Platypus](https://huggingface.co/datasets/garage-bAInd/Open-Platypus). |
|
|
|
- **Developed by:** STEM.AI |
|
- **Model type:** Q&A and code generation |
|
- **Language(s) (NLP):** English |
|
- **Finetuned from model:** [microsoft/phi-2](https://huggingface.co/microsoft/phi-2) |
|
|
|
|
|
### Direct Use |
|
|
|
Q&A related to electrical engineering, and Kicad software. Creation of Python code in general, and for Kicad's scripting console. |
|
|
|
Refer to [microsoft/phi-2](https://huggingface.co/microsoft/phi-2) model card for recommended prompt format. |
|
|
|
### Inference script |
|
|
|
[Standard](https://github.com/STEM-ai/Phi-2/blob/4eaa6aaa2679427a810ace5a061b9c951942d66a/chat.py) |
|
|
|
[GPTQ format](https://github.com/STEM-ai/Phi-2/blob/ab1ced8d7922765344d824acf1924df99606b4fc/chat-GPTQ.py) |
|
|
|
## Training Details |
|
|
|
### Training Data |
|
|
|
Dataset related to electrical engineering: [STEM-AI-mtl/Electrical-engineering](https://huggingface.co/datasets/STEM-AI-mtl/Electrical-engineering) |
|
It is composed of queries, 65% about general electrical engineering, 25% about Kicad (EDA software) and 10% about Python code for Kicad's scripting console. |
|
|
|
In additionataset related to STEM and NLP: [garage-bAInd/Open-Platypus](https://huggingface.co/datasets/garage-bAInd/Open-Platypus) |
|
|
|
### Training Procedure |
|
[LoRa script](https://github.com/STEM-ai/Phi-2/blob/4eaa6aaa2679427a810ace5a061b9c951942d66a/LoRa.py) |
|
|
|
A LoRa PEFT was performed on a 48 Gb A40 Nvidia GPU. |
|
|
|
## Model Card Authors |
|
|
|
STEM.AI: [email protected]\ |
|
[William Harbec](https://www.linkedin.com/in/william-harbec-56a262248/) |
|
|
|
|
