CodeLlamas1

sentence splitting for llama
setting up llama
data uploaded for Llama2 to compute on successfully
llama response

Inspiration

We picked this project at the datathon because we want to solve the problem of unexpected traffic jams, known as 'ghost' traffic. By using data and LLama, we aim to understand why these jams happen and find ways to prevent them. Our goal is to make driving smoother and easier for everyone.

What it does

Our project uses the state of the art LLAMA model to provide insights into the Traffic congestion (2016-2022) data. It helps us determine how features Time of Day Effects, Day of Week Effects, Seasonal Effects, Distance from Start Point, Time of Departure, Traffic Incidents & Road Network are the critical factors that can help us build our model that will in turn help drivers avoid "Ghost traffic".

How we built it

We use langchaIn which allows us to customize LLAMA based on our data and Replicate which runs machine learning models in the cloud. And then, we provide several prompts to help us gain insight on the data. Also, we have created a prediction model to predict severity based on all other factors, we planned on improving this model with the data analytics insights that we gained from llama 2's analysis.

Challenges we ran into

The biggest challenge was figuring out how to use Llama 2. Since Llama 2 is a relatively new LLM, we had trouble finding resources online to help us build our model faster. Other than that we had trouble finding a right platform for using Llama considering the computing power required for computation on a file of this size would significantly increase the need for more capable computing power.

Accomplishments that we're proud of

We are proud of the fact that despite all these challenges, we kept on working on the project till the very end and were able to achieve some milestones along the way. Some of these include-

1) Setting up a model to predict severity - which according to our team's plan would serve as a template to modify variables after finding more insights from the Llama 2 LLM. 2) Setting up Llama 2 successfully and incorporating our data, allowing the LLM to give us useful insights on important variables affecting the flow of traffic 3) Overall we are proud of our approach of breaking down bigger, complex tasks into smaller easier tasks and asking for help once we exhausted our available resources. Above all, we are glad we didn't give up and stuck with the challenge till the very end.

What we learned

We gained valuable experience in using LLama 2 for analyzing traffic data and extracting insights to tackle unexpected traffic jams. Additionally, we improved our problem-solving skills and resilience in overcoming challenges!

What's next for CodeLlamas

Moving forward, we plan to further refine our model using the insights generated from LLama 2. We aim to explore additional machine learning techniques and optimize our solution to contribute towards smoother and more efficient traffic flow.