lora-adapters

Local-first AI citizens on your own hardware. Teams of continuously-learning personas (Rust core, llama.cpp fork, LoRA genomes) that collaborate, remember, dream, and improve on public benchmarks — on a MacBook. Not assistants you rent: minds you host.

Open-source framework for turning expert knowledge into PII-free synthetic conversational data and production-ready LoRA adapters.

SOVL System (Self-Organizing Virtual Lifeform): A complex, purpose-agnostic autonomous agent with continuous, asynchronous learning capabilities via a dynamic scaffolded LLM and a frozen base LLM

Custom trl.SFTTrainer that adds a KL divergence loss between a LoRA-adapted model and its base model.

A Python-based LoRa adapter that bridges ChirpStack MQTT messages to SuperMQ NATS, providing seamless integration between LoRaWAN devices and the SuperMQ IoT platform.

Build an end-to-end LoRA fine-tuning pipeline for a 4-bit Qwen2.5-0.5B chat model using Unsloth. You'll load the quantized base, attach LoRA adapters, format a tiny instruction dataset, run a featherweight SFT job, and generate a reply from the tuned model.

Production-grade LLM inference platform with dynamic LoRA swapping. Serves REST APIs for Llama/Mistral/Qwen. Scaled on Kubernetes; monitored via Prometheus/Grafana. Staff-level ML platform artifact

The goal is to study how different dataset combinations affect model performance.

This project allows you to fine-tune LLaMA 3.2B Instruct using LoRA adapters on CPU.

Fine-tuned Llama 3.1 8B model for cancer-specific entity extraction (IE) serving using vLLM on EC2 instance with production backend and frontend