What are the hardware requirements for Japanese GPT Neox 3.6B?

Japanese GPT Neox 3.6B requires approximately 7.4 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 Japanese GPT Neox 3.6B and how large is it?

Japanese GPT Neox 3.6B is developed by rinna, a model with 3.6b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Japanese GPT Neox 3.6B perform on standard benchmarks?

Japanese GPT Neox 3.6B has the following published scores: MMLU 23.12. Compare against reference models on this page or on the LLM Explorer leaderboards.

Japanese GPT Neox 3.6B by rinna — VRAM 7.4GB, 2K context

Name: Japanese GPT Neox 3.6B
Author: rinna

Japanese GPT Neox 3.6B is an open-source language model by rinna. Features: 3.6b LLM, VRAM: 7.4GB, Context: 2K, License: mit, LLM Explorer Score: 0.15, Arc: 25, HellaSwag: 25.5, MMLU: 23.1.

Arxiv:2404.01657 Autotrain compatible Dataset:cc100 Dataset:mc4 Dataset:wikipedia Gpt neox Ja Lm Pytorch Region:us Safetensors

Model Card on HF 🤗: https://huggingface.co/rinna/japanese-gpt-neox-3.6b

Japanese GPT Neox 3.6B Parameters and Internals

Model Type

text generation, nlp

Additional Notes

The model uses sentencepiece tokenizer with specific configurations such as byte fallback and `-add_dummy_prefix` turned off.

Training Details

Data Sources:

Japanese CC-100, Japanese C4, Japanese Wikipedia

Data Volume:

312.5B tokens

Methodology:

traditional language modelling objective

Model Architecture:

A 36-layer, 2816-hidden-size transformer-based language model

Input Output

Performance Tips:

set use_fast=False to make certain features function correctly.

LLM Name	Japanese GPT Neox 3.6B
Repository 🤗	https://huggingface.co/rinna/japanese-gpt-neox-3.6b
Model Size	3.6b
Required VRAM	7.4 GB
Updated	2025-09-23
Maintainer	rinna
Model Type	gpt_neox
Model Files	7.4 GB 7.4 GB
Supported Languages	ja
Model Architecture	GPTNeoXForCausalLM
License	mit
Context Length	2048
Model Max Length	2048
Tokenizer Class	T5Tokenizer
Padding Token	[PAD]
Vocabulary Size	32000
Torch Data Type	float16

Rank the Japanese GPT Neox 3.6B 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 Japanese GPT Neox 3.6B

Best Alternatives	Context / RAM	Downloads	Likes
...y Jimba Japanese Large Lm 3.6B	2K / 7.1 GB	6	0
...rrowSmartPlus 3.6B Instruction	2K / 14.3 GB	9	1
...rrowSmartPlus 3.6B Instant Sft	2K / 14.3 GB	7	1
...rtPlus 3.6B Instant Sft JHSVer	2K / 14.3 GB	9	1
... GPT Neox 3.6B Instruction Sft	2K / 7.4 GB	9602	105
... GPT Neox 3.6B Instruction Ppo	2K / 7.4 GB	3052	74
... Large Lm 3.6B Instruction Sft	2K / 7.2 GB	127	27
Japanese Large Lm 3.6B	2K / 7.2 GB	152	73
...T Neox 3.6B Instruction Sft V2	2K / 7.4 GB	627	26
...tion Sft 8bit 1g Actorder True	2K / 2.8 GB	51	3