AWS Machine Learning Blog:Fine-tune Code Llama on Amazon SageMaker JumpStart [*]

Wrote about HumanEval and CodeLlama Reasoning.

How to deploy LLaVA Models to SageMaker Endpoints [*]

Showed how you can deploy any VLM to Amazon SageMaker.

Magikarp: Applying RL Policies to Code Generation Models
[model][dataset]

Reproduced the Magicoder paper results using CodeLlama 7B, Mixtral for synthetic data generation, and Axolotl for fine-tuning. I employed reinforcement learning with OpenAI Gym and LocalStack to create a policy that enforces AWS-specific correctness for the awscli and CloudFormation. This approach ensures the generated code adheres to AWS best practices and API requirements while solving the given task.

Nuvo: A Friend For Any City
[project]

Built during the MongoDB SF AI Hackathon w TogetherAI and Langchain (and won!). Created a search engine that lets you ask Anthony Bourdain for restaurant recommendations near you. Attempting to integrate collaborative filtering, image preference embeddings, and more. Still building this but you can try it today!

Salesparty: Prospecting At Scale For AWS Sellers

Built a research agent internally at Amazon! Helped people save hours of time and I learned how to ship an entire mini-product at a big company!