Skip to content

Setup guides / from first model to a working system

Build a baseline you can trust.

Practical paths through hardware, runtimes, memory, routing, and measurement. Based on the lab’s configuration and the mistakes its experiments exposed.

New to local inference?

Start with one useful workload.

Choose an artifact, budget memory, and qualify one backend before building a fleet.

9 OF 9 GUIDES

01 / Fundamentals

Your first useful local LLM

Start with one workload, one model, and a memory budget. Expand after a measured baseline.

2 MIN READ / UPDATED OCT 2026

02 / GB10

DGX Spark & GX10: capacity with discipline

Our DGX Spark and ASUS GX10 now run DeepSeek through TensorFold TP2. Understand unified memory, four shared slots, model staging and measured limits.

3 MIN READ / UPDATED OCT 2026

03 / RTX 5090

RTX 5090 on Windows + WSL2

Verify the Windows driver from Linux, enable CUDA, and budget a 32 GB GPU for useful work.

2 MIN READ / UPDATED OCT 2026

04 / Mac Studio

Mac Studio: MLX, Metal & memory ownership

Run local models on Apple silicon without confusing installed memory, configured context, and verified capacity.

2 MIN READ / UPDATED OCT 2026

05 / Operations

Routing is easy. Model lifecycle is the work.

Persistence, swap groups, alias drift, child processes, and protocol fidelity across heterogeneous machines.

2 MIN READ / UPDATED OCT 2026

06 / Measurement

Measure the experience, not just tok/s

A repeatable protocol for latency, decode, context, concurrency, correctness, and honest comparisons.

2 MIN READ / UPDATED OCT 2026

07 / Operations

The problems behind a model that “works”

Lost tool calls, memory that survives unloading, untested context, bad gauges, and regressions.

2 MIN READ / UPDATED OCT 2026

08 / Operations

Operating a coder across two GB10 hosts

Operate the current DeepSeek TensorFold pair: shared admission, model identity, Engram staging, memory guards, readiness, rollback and pending maintenance.

5 MIN READ / UPDATED OCT 2026

09 / Deployment

DeepSeek on two GB10 hosts: stage, validate, compare

The October 23 DeepSeek deployment, from verified checkpoint and Engram staging to four shared slots, a matched GLM control and reversible operations.

8 MIN READ / UPDATED OCT 2026