Tensor Wiki · direct technical answers

Find the file.
Then decide.

Short, source-linked explanations for the questions that usually appear five minutes before someone deletes the wrong model folder.

Operational facts firstVersion-sensitive claims sourcedPlain-language boundaries
A modern path map connecting local model folders, caches, formats and runtime locations.

Browse the Wiki

Start with the storage you can see.

These pages help locate and understand existing local data. They do not tell you to delete a cache or archive until you know what depends on it.

LoRA

What is LoRA in AI? Low-Rank Adaptation explained

See why a small learned adapter can steer a frozen base model—and why that adapter still depends on the base.

Format

What is GGUF? Tensors, metadata and quantization

Understand the inference-oriented container used by llama.cpp without assuming that every GGUF is quantized.

Format

GGUF vs. SafeTensors: choose by workflow

Compare runtime purpose, packaging, metadata and conversion boundaries instead of treating extensions as quality tiers.

Artifacts

Checkpoint vs. LoRA: complete state or adapter update?

Know what can load as the main model, what requires a base and which related artifacts make recovery complete.

LoRA

What is a LoRA file?

Inspect adapter weights, SafeTensors packaging, companion configuration and the exact base-model dependency.

Cache

Change the Hugging Face cache directory

Choose HF_HOME or HF_HUB_CACHE, migrate existing snapshots and prove the new path before cleanup.

Runtime

Ollama change model directory: move and verify safely

Set OLLAMA_MODELS on macOS, Linux or Windows and keep the old store until a known model survives restart.

Runtime

ComfyUI models folder: native and extra model paths

Place checkpoints and LoRAs in the right native folders, or connect one external shared model library without making another copy.

Cache

Where is the Hugging Face cache directory?

Find the Hugging Face cache location, inspect overrides, and distinguish a cache snapshot from an intentional model library.

Runtime

Ollama model location: where are models stored?

Find the Ollama models directory, check OLLAMA_MODELS, and avoid moving a live runtime folder blindly.

Training

LoRA vs. QLoRA: what changes and what stays?

Compare QLoRA vs. LoRA when deciding what a training artifact represents and what must remain reproducible.

Format

What is a SafeTensors file?

Understand the SafeTensors file format, what “safe” means, and why an extension is not a full retention or trust policy.

Runtime

LM Studio model location: where does LM Studio store models?

Use LM Studio’s own model inventory and import workflow before changing local model storage.