|
--- |
|
license: mit |
|
datasets: |
|
- ashmib/wikivoyage-eu-cities-qa |
|
language: |
|
- en |
|
base_model: |
|
- google/gemma-2-2b-it |
|
pipeline_tag: question-answering |
|
--- |
|
|
|
# Finetuning Gemma-2b-it Models to answer European travel-related Q&A |
|
|
|
This is a sample project where we fine-tuned the Gemma-2b-it model on Wikivoyage travel data on European cities to see if it answers the travel related Q&A better. |
|
This is also an ongoing WIP project. |
|
|
|
### Training Specs |
|
* Gemma-2b fine-tuned on 2 Nvidia A40 GPUs using HuggingFace SFTrainer |
|
* 10 epochs |
|
* 1e-5 learning rate |
|
|
|
### Data |
|
* [Wikivoyage (English) data](https://huggingface.co/datasets/ashmib/wikivoyage-eu-cities-qa) for 160 European cities. |
|
* Used Gemini 1.5 Pro and Flash to extract relevant questions from the data. |
|
|
|
[Code](https://github.com/ashmibanerjee/eu-travel-reco) | [Demo - TBD](https://huggingface.co/ashmib) |
