Neurolume

Neurolume is a local desktop application that fine-tunes language models and visualizes neural network activity.

Neurolume is a local app designed for fine-tuning large language models and inspecting their internal operations layer by layer. Users can fine-tune models using custom datasets, chat histories, or notes, and review automatic data checks. The interface provides a live 3D view of model activity, layer weight maps, a logit lens to inspect intermediate predictions, and an answer replay tool showing token confidence. It supports exporting trained adapters in MLX and Ollama formats, chatting with an OpenAI-compatible local server, importing documents for retrieval, and using web search and an agent for project edits.

Neurolume at a glance

Best for: Software engineer, Researcher
Runs on: macOS, Web, API, self-hosted, open source
Pricing: Free
Released: November 2026
Category: AI coding tool
Website: github.com/yiteng-chen/neurolume

Key features

Fine-tune language models locally with guided LoRA training
View 3D visualizations of layer activity during generation and training
Inspect weight matrices and layer changes with the layer map
Analyze intermediate predictions using the logit lens
Replay answers with per-token confidence tracking
Export trained models to MLX and Ollama formats
Import documents for knowledge base retrieval
Run an OpenAI-compatible local server with web search support

Use it for

Fine-tune language models on custom data and chat history
Monitor neural network layer activity and weights in real time
Analyze token confidence and alternative predictions for model outputs
Export and deploy trained model adapters to MLX or Ollama

Alternatives to Neurolume

AirLLM — Runs massive 70B+ parameter models on small 4GB consumer GPUs.
Overview — Learn language models by running hands-on experiments without writing scripts
Soup — Fine-tune and align models like Llama-3.1-8B on a 4 GB laptop GPU.
LlamaFactory — Provides a code-free CLI and Web UI for training 100+ different LLMs.
LMFlow — It offers memory-efficient training and full Accelerate support for large language models.
WebLLM — Runs high-performance language models directly inside web browsers.
llm-d — Deploys production-grade LLM serving on existing hardware clusters.
Ludwig — Train and fine-tune complex AI models using a single YAML configuration file.
MLC LLM — Enables native optimization and deployment of AI models across platforms.
MLX LM — Runs LLMs and fine-tuning natively on Apple silicon hardware.
Argilla — Provides programmatic tools for continuous data evaluation and human-in-the-loop feedback.
MNN — It delivers high-performance inference and training with a lightweight footprint.

All Neurolume alternatives