What are the hardware requirements for GPT NeoX 20B Erebus?

GPT NeoX 20B Erebus requires approximately 41.4 GB of VRAM and supports a context window of 2K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed GPT NeoX 20B Erebus and how large is it?

GPT NeoX 20B Erebus is developed by KoboldAI, a model with 20b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does GPT NeoX 20B Erebus perform on standard benchmarks?

GPT NeoX 20B Erebus has the following published scores: MMLU 26.77. Compare against reference models on this page or on the LLM Explorer leaderboards.

GPT NeoX 20B Erebus by KoboldAI — VRAM 41.4GB, 2K context

Name: GPT NeoX 20B Erebus
Author: KoboldAI

GPT NeoX 20B Erebus is an open-source language model by KoboldAI. Features: 20b LLM, VRAM: 41.4GB, Context: 2K, License: apache-2.0, LLM Explorer Score: 0.07, Arc: 45.5, HellaSwag: 72.8, MMLU: 26.8, GSM8K: 2.3.

Arxiv:2204.06745 En Gpt neox Pytorch Region:us Sharded

Model Card on HF 🤗: https://huggingface.co/KoboldAI/GPT-NeoX-20B-Erebus

GPT NeoX 20B Erebus Parameters and Internals

Model Type

autoregressive, language model

Use Cases

Limitations:

Strong NSFW bias

Additional Notes

Warning: THIS model is NOT suitable for use by minors. The model will output X-rated content.

Training Details

Data Sources:

Literotica, Sexstories, Dataset-G, Doc's Lab, Pike Dataset, SoFurry

Methodology:

Heavily modified version of Ben Wang's Mesh Transformer JAX library

Hardware Used:

TPUv3-256 TPU pod

Responsible Ai Considerations

Mitigation Strategies:

Warning: This model has a very strong NSFW bias!

LLM Name	GPT NeoX 20B Erebus
Repository 🤗	https://huggingface.co/KoboldAI/GPT-NeoX-20B-Erebus
Model Size	20b
Required VRAM	41.4 GB
Updated	2026-07-31
Maintainer	KoboldAI
Model Type	gpt_neox
Model Files	2.0 GB: 1-of-23 2.0 GB: 2-of-23 1.9 GB: 3-of-23 1.8 GB: 4-of-23 1.9 GB: 5-of-23 1.8 GB: 6-of-23 1.8 GB: 7-of-23 1.9 GB: 8-of-23 2.0 GB: 9-of-23 1.8 GB: 10-of-23 1.9 GB: 11-of-23 2.0 GB: 12-of-23 1.8 GB: 13-of-23 1.8 GB: 14-of-23 2.0 GB: 15-of-23 1.9 GB: 16-of-23 2.0 GB: 17-of-23 1.4 GB: 18-of-23 1.8 GB: 19-of-23 1.8 GB: 20-of-23 2.0 GB: 21-of-23 1.8 GB: 22-of-23 0.3 GB: 23-of-23
Supported Languages	en
Model Architecture	GPTNeoXForCausalLM
License	apache-2.0
Context Length	2048
Model Max Length	2048
Transformers Version	4.19.0.dev0
Tokenizer Class	GPTNeoXTokenizer
Vocabulary Size	50432
Torch Data Type	float16

Rank the GPT NeoX 20B Erebus capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to GPT NeoX 20B Erebus

Best Alternatives	Context / RAM	Downloads	Likes
GPT Neox 20B	2K / 40.8 GB	794665	583
GPT NeoXT Chat Base 20B	2K / 41.2 GB	256	694
EleutherAI GPT Neox 20B 4bits	2K / 12.5 GB	8	0
H2ogpt Oasst1 512 20B	2K / 41.2 GB	1064	39
...t Gm Oasst1 Multilang 1024 20B	2K / 41.2 GB	140	10
H2ogpt Gm Oasst1 En 1024 20B	2K / 41.2 GB	55	4
GPT Neox 20B Full Precision	2K / 82.5 GB	52	0
Oasst GPT Neox 20B 1000 Steps	2K / 41.2 GB	75	0
Oasst GPT Neox 20B 3000 Steps	2K / 41.2 GB	60	0
GPTNeoX 20B TestGen Dart V1.0	2K / 41.2 GB	5	2

Note: green Score (e.g. "73.2") means that the model is better than KoboldAI/GPT-NeoX-20B-Erebus.