Advanced Data Intelligence

Distilled intelligence for local machines.

Compact ADI models distilled from frontier teachers, quantized to GGUF, and built at theLAB for Ollama, llama.cpp, and private local inference.

theLAB source HF models Live demo Run local

adi.local/distill

teacher: glm-5.2

student: qwen3.5-4b

format: gguf:q4_k_m

runtime: ollama

Model rail

Pick a student. Copy a command. Run offline.

general assistant

adi-qwen3.5-4b-glm5.2-general

The smallest flagship ADI general model: distilled from glm-5.2 for practical local reasoning, explanation, and tool-aware chat.

ollama run hf.co/AdvancedDataIntelligence/adi-qwen3.5-4b-glm5.2-general-GGUF:Q4_K_M

Open model card

Live browser demo Try adi-qwen3.5-4b-glm5.2-general in Hugging Face Spaces.

Local-first

Not every intelligence should require an API call.

ADI is built for privacy, offline reliability, reproducible tools, and consumer hardware. Local models keep useful intelligence close to the machine doing the work: available when the network is not, inspectable when the stack matters, and practical without renting a remote brain for every thought.

01Seed prompts
02Teacher answers
03Student tune
04GGUF
05Ollama
06Published

Pipeline

Teacher answers become local weights.

ADI is not a prompt wrapper. It is a repeatable distillation workflow: seed prompts, teacher responses, student fine-tune, adapter merge, GGUF conversion, local runtime, and public model cards.

View dataset