|
--- |
|
language: |
|
- en |
|
library_name: transformers |
|
tags: |
|
- physics |
|
- lora |
|
- Qlora |
|
- phi-2 |
|
datasets: |
|
- ayoubkirouane/arxiv-physics |
|
pipeline_tag: text-generation |
|
--- |
|
|
|
|
|
## Phi-2-arxiv-physics : |
|
|
|
Phi-2-arxiv-physics is a small language model (SLM) derived from Microsoft Research's Phi-2, a Transformer-based model with 2.7 billion parameters. |
|
Specifically fine-tuned using the innovative QLoRA (Quantized Low-rank Adapters) technique, this model exhibits remarkable capabilities in the domain of physics-related text. |
|
|
|
|
|
|
|
|
|
## Fine-Tuning: |
|
|
|
The model has been fine-tuned using **QLoRA**, a technique that reduces computational and memory requirements during the fine-tuning process. |
|
|
|
This two-stage approach involves quantizing parameters and introducing learnable low-rank adapters, resulting in faster fine-tuning with reduced memory demands. |
|
|
|
## Data Source: |
|
|
|
**Phi-2-arxiv-physics** is specifically trained on the **arXiv-physics** dataset, consisting of questions and answers related to physics. |
|
|
|
This dataset is sourced from the arXiv repository, encompassing a wide range of topics within the field of physics. |
|
|
|
|
|
Usage (High-level) : |
|
``` |
|
!pip install -q -U git+https://github.com/huggingface/transformers.git |
|
|
|
``` |
|
|
|
``` python |
|
|
|
from transformers import pipeline |
|
|
|
pipe = pipeline("text-generation", model="ayoubkirouane/phi-2-arxiv-physics") |
|
|
|
``` |
|
|
