Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

Model weights

Model weights are the learned numbers inside a neural network, set during training, that decide how it turns input into output. A checkpoint is a saved file or set of files containing those weights.

Also called: weights, checkpoint, parameters file

By Divya Sharma · Checked Nov 15, 2026 · How we review

A neural network is a fixed structure of layers, and the weights are the values in those layers. Training adjusts them until the model's outputs are good. When people say a lab "released the weights", they mean you can download those numbers and run the model yourself. The count of weights is the model's parameter count.

What a checkpoint contains

In PyTorch, the weights live in a state_dict, a dictionary that maps each layer to its tensor. A checkpoint saves that, sometimes with optimizer state so training can resume. Model files on Hugging Face are usually split into several shards and come with a config file describing the architecture and a tokenizer. Without the matching architecture code and tokenizer, the weights alone will not run.

Formats

Common formats include PyTorch's pickle-based files, Safetensors, GGUF for local runners such as llama.cpp, and ONNX. The format affects safety, loading speed and which tools can read it.

Size and precision

The file size is roughly the parameter count times the bytes per weight. Storing weights in lower precision or applying quantization shrinks the file and the memory needed to run it, with some loss of quality.

Weights vs open source

Having the weights is not the same as having everything. An open-weight model may come without its training data or code, and under a licence with limits. The Open Source Initiative's definition treats parameters as only one of the parts that must be shared.

Related terms

Open-weight model
An open-weight model is a model whose trained weights are published for anyone to download and run, often without the training code or full training data, so it is not necessarily open source.
Safetensors
Safetensors is a file format from Hugging Face for storing model weights. Unlike Python pickle files, loading it cannot run hidden code, and it supports fast, memory-mapped loading of individual tensors.
Model parameters
Model parameters are the numbers a neural network learns in training, mostly weights. The parameter count, written as 7B for seven billion, is the usual shorthand for a model's size.
GGUF
GGUF is a binary file format for storing a model's weights, tokenizer and metadata in one file, built for fast loading and local inference with GGML-based tools such as llama.cpp, often in quantized form.
Numeric precision (FP16, BF16, FP8)
Numeric precision is the number format used to store a model's weights and run its maths, such as FP32, FP16, BF16, FP8 or INT4. Fewer bits per number means less memory and faster inference, at some risk to accuracy.
Model license
A model license is the legal terms under which a model's weights, and sometimes its code and data, may be used, changed and shared. Terms range from permissive open licences to non-commercial or custom restricted ones.

Sources