Understand the model before you deploy it.
A source-first technical knowledge base for model weights, hardware, formats, runtimes, RAG, fine-tuning, licensing and European deployment.
Each guide is written answer-first for search and generative retrieval, then expands into architecture, trade-offs, decision rules, diagrams and primary sources.
Start with the question you need answered.
The knowledge base complements the model registry: concepts explain how to decide; Model Passports show how those concepts apply to concrete checkpoints.
What Are Open-Weight AI Models?
Definition, examples, weight access and why open-weight deployment matters.
Read reference →Open Weight vs Open Source AI vs Proprietary Models
Separate weight availability, source openness, licensing and hosted-only access.
Read reference →Open-Weight Models vs API Models
Compare control, cost structure, privacy, operations and deployment trade-offs.
Read reference →AI Model Parameters Explained
Understand 7B, 70B, total parameters, active parameters and model scale.
Read reference →Mixture of Experts (MoE) Explained
Understand sparse routing, total vs active parameters and deployment implications.
Read reference →LLM Quantization Explained
FP16, FP8, INT8, INT4, GGUF, AWQ and GPTQ without the marketing shorthand.
Read reference →How Much RAM and VRAM Do Open-Weight Models Need?
Estimate model-memory requirements and understand KV cache, context and runtime overhead.
Read reference →GGUF vs Safetensors
Choose the right model-weight format for local inference, training and serving.
Read reference →Ollama vs llama.cpp vs vLLM vs SGLang
Compare local UX, low-level inference and production serving runtimes.
Read reference →How to Run Open-Weight Models Locally
A practical path from model choice and quantization to local and self-hosted inference.
Read reference →RAG with Open-Weight Models
Retrieval, embeddings, vector stores, permissions and private knowledge workflows.
Read reference →Fine-Tuning Open-Weight Models: LoRA, QLoRA and Full Fine-Tuning
Choose the right adaptation method by data, memory, risk and deployment needs.
Read reference →Open-Weight Model Licenses Explained
Apache 2.0, MIT, community licenses, custom terms and commercial-use checks.
Read reference →Open-Weight AI in Europe: GDPR, Data Residency and the AI Act
Separate provider origin, EU deployment, data residency, GDPR and AI Act roles.
Read reference →How to Choose an Open-Weight Model in 2026
A technical decision framework for workload, hardware, license, runtime and governance.
Read reference →Concept → model → deployment.
Use a concept guide to understand the decision, then move into the 64-model directory and the EU deployment layer for model-specific evidence.
From understanding to model choice.
The comparison hub applies the concepts in this knowledge base to concrete hardware, licensing, architecture and workload shortlists.