Llama Impact Hackathon

Start

Llama 3 finetune finder for specific use cases.

Llama finder

Luis Guerra

Small ML models can compete and even outperform larger and powerful ones when fine-tuned for specific tasks. This implies higher throughput, lower latency and lower costs. Additionally, it could enable deployment of ML models on edge devices.

Unlike Chatbot Arena which compares LLMs in general benchmarks, Llama Finder’s goal is to help you pick the right Llama finetuned for your specific use-case.

Llama 3 finetune finder takes as input prompt your specific needs. It then queries HuggingFace looking for specific keywords. It summarizes the available models with Llama 3 using Together AI API, and verifies if they are readily available in Featherless.ai. 
It also uses Brave API to crawl for web resources. In the future, image search could be added and analyzed with Llama 3.2 vision capabilities. Alternatively, I intend to use DSPy to do multihop reasoning. At the moment "Llama Finder" is only able to summarize resources but it's not capable of doing a higher level comparison between the different Llama candidates. It also lists candidates based on alternative open source models.
The comparison table provides a summary of the model, whether it's finetuned to follow instructions, parameters and specific datasets used to fine-tune the model. It also listed resources crawled from the web at the bottom.

The comparison table provides a summary of the model, whether it's finetuned to follow instructions, parameters and specific datasets used to fine-tune the model. It also listed resources crawled from the web at the bottom.

Llama 3 finetune finder. AI assitant to crawl HuggingFace and the web, in order to help you find the right Llama finetune for your specific needs.

"Llama finder team on Llama Impact Hackathon Hackathon"

Team Idea

Submission