GPT4All

2.8.2 verified Fri May 01 auth: no python

Python bindings for GPT4All, a locally running open-source LLM ecosystem. Version 2.8.2 supports local inference with models like Mistral, Llama, and GPT4All. Released regularly with updates to model support and API improvements.

pip install gpt4all

Common errors

error ValueError: Unknown model name 'my-model' ↓

cause Trying to instantiate a model with an incorrect or unsupported name.

fix

Call GPT4All.list_models() to see available model names, or provide a valid file path using model_path.

error ModuleNotFoundError: No module named 'gpt4all' ↓

cause The package is not installed or installed in a different environment.

fix

Run pip install gpt4all in your active Python environment.

error AttributeError: 'GPT4All' object has no attribute 'generate' ↓

cause Using an older version of the library or a model that does not support text generation.

fix

Update to the latest version with pip install --upgrade gpt4all. Ensure the model is a text generation model (not an embedding model).

Warnings

breaking In version 2.5.0+, the GPT4All class constructor changed to require a model name string instead of a model object. Old code using `GPT4All(llmodel=...)` will break. ↓

fix Use `GPT4All('model-name')` or `GPT4All(model_path='/path/to/model.bin')`.

gotcha Model names in GPT4All v2.x must match the exact name in the official model list (e.g., 'Meta-Llama-3-8B-Instruct-4bit'). Using a wrong name downloads a new model or raises an exception. ↓

fix Check available models with `GPT4All.list_models()` and use the exact `name` field.

deprecated The `generate()` method's `n_predict` parameter was renamed to `max_tokens` in v2.7.0. Old `n_predict` still works but is deprecated and will be removed. ↓

fix Use `max_tokens` instead of `n_predict`.

gotcha By default, models are downloaded to the user's cache directory (~/.cache/gpt4all/). If you expect a model to be in the current directory, specify `model_path` explicitly. ↓

fix Use `GPT4All(model_path='./model.bin')` to load from current directory.

Imports

GPT4All
wrong
```
from gpt4all.gpt4all import GPT4All
```
correct
```
from gpt4all import GPT4All
```
Older versions exposed the class under a different submodule path.

Quickstart

Initialize model with a known model name (or path to local file), then generate text.

from gpt4all import GPT4All
model = GPT4All("Meta-Llama-3-8B-Instruct-4bit")
output = model.generate("What is the capital of France?", max_tokens=50)
print(output)