What are the hardware requirements for Pythia 70M?

Pythia 70M requires approximately 0.2 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 Pythia 70M and how large is it?

Pythia 70M is developed by EleutherAI, a model with 70m parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Pythia 70M perform on standard benchmarks?

Pythia 70M has the following published scores: MMLU 25.9. Compare against reference models on this page or on the LLM Explorer leaderboards.

Pythia 70M by EleutherAI — VRAM 0.2GB, 2K context

Name: Pythia 70M
Author: EleutherAI

Pythia 70M is an open-source language model by EleutherAI. Features: 70m LLM, VRAM: 0.2GB, Context: 2K, License: apache-2.0, LLM Explorer Score: 0.19, Arc: 21.6, HellaSwag: 27.3, MMLU: 25.9, GSM8K: 0.3.

Arxiv:2101.00027 Arxiv:2201.07311 Arxiv:2304.01373 Dataset:eleutherai/pile Deploy:azure En Gpt-neox Gpt neox Pythia Pytorch Region:us Safetensors

Model Card on HF 🤗: https://huggingface.co/EleutherAI/pythia-70m

Pythia 70M Parameters and Internals

Model Type

Transformer-based Language Model, Causal Language Model

Use Cases

Areas:

Research

Primary Use Cases:

Research on large language model behavior and interpretability

Limitations:

Not suitable for deployment or human-facing interactions, Limited to English language only

Additional Notes

Suit designed to promote large language model research.

Supported Languages

English (Fluent)

Training Details

Data Sources:

The Pile

Data Volume:

299,892,736,000 tokens

Methodology:

Uniform batch size, Flash Attention, checkpoints at steps {0,1,2,4...143000}.

Model Architecture:

Transformer

Input Output

Input Format:

Text input

Accepted Modalities:

text

Output Format:

Text output

Release Notes

Version:

Current

Notes:

Included checkpoints, uniform batch size, Flash Attention.

LLM Name	Pythia 70M
Repository 🤗	https://huggingface.co/EleutherAI/pythia-70m
Model Size	70m
Required VRAM	0.2 GB
Updated	2026-08-01
Maintainer	EleutherAI
Model Type	gpt_neox
Model Files	0.2 GB 0.2 GB
Supported Languages	en
Model Architecture	GPTNeoXForCausalLM
License	apache-2.0
Context Length	2048
Model Max Length	2048
Transformers Version	4.24.0
Tokenizer Class	GPTNeoXTokenizer
Vocabulary Size	50304
Torch Data Type	float16

Rank the Pythia 70M 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 Pythia 70M

Best Alternatives	Context / RAM	Downloads	Likes
Pythia 70M Deduped	2K / 0.2 GB	677080	29
Pythia 70m Sft Hh	2K / 0.1 GB	9	0
Pythia 70M Multi	2K / 0.3 GB	6	0
Pythia Chinese 70M UDN	2K / 0.2 GB	18	0
Pythia 70m Sft	2K / 0.3 GB	7	0
...ythia 70M Wikipedia Paragraphs	2K / 0.3 GB	4	3
...a 70M Deduped Cleansharegpt En	2K / 0.3 GB	52	0
...thia 70M Deduped Cleansharegpt	2K / 0.3 GB	138	1
Chessdevilai	2K / 0.3 GB	5	0
...ythia 70M Deduped Step44k 92bt	2K / 0.2 GB	113	0

Note: green Score (e.g. "73.2") means that the model is better than EleutherAI/pythia-70m.