---
library_name: mlx
tags:
- imagegpt
- vision
- mlx
- safetensors
- bfloat16
- icl-many-replication
datasets:
- imagenet-21k
base_model: openai/imagegpt-large
---

# ImageGPT Large BF16 MLX

This repository contains the local bfloat16/MLX conversion of
[`openai/imagegpt-large`](https://huggingface.co/openai/imagegpt-large) used by
the ImageGPT runs in the ICLManyReplication artifact for "Many Next-Token
Predictors are In-Context Learners".

The upload includes:

- `model.safetensors`: converted bfloat16 model weights
- `config.json`: ImageGPT model configuration with the local MLX model file hook
- `imagegpt_mlx_lm.py`: local MLX model implementation used by the replication harness
- `preprocessor_config.json`: original ImageGPT preprocessing configuration
- `conversion_metadata.json`: source path, dtype, tensor count, and conversion notes

The conversion metadata records `openai/imagegpt-large` weights converted from
the original PyTorch checkpoint into a bfloat16 safetensors artifact.

## Intended Use

This checkpoint is intended as a drop-in artifact for reproducing the ImageGPT
principal-result row in ICLManyReplication. For general ImageGPT usage, see the
upstream [`openai/imagegpt-large`](https://huggingface.co/openai/imagegpt-large)
model card.