What are the hardware requirements for Smol Llama 220M GQA?

Smol Llama 220M GQA requires approximately 0.4 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 Smol Llama 220M GQA and how large is it?

Smol Llama 220M GQA is developed by BEE-spoke-data, a model with 220m parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Smol Llama 220M GQA perform on standard benchmarks?

Smol Llama 220M GQA has the following published scores: MMLU 25.85. Compare against reference models on this page or on the LLM Explorer leaderboards.

Smol Llama 220M GQA by BEE-spoke-data — VRAM 0.4GB, 2K context

Name: Smol Llama 220M GQA
Author: BEE-spoke-data

Smol Llama 220M GQA is an open-source language model by BEE-spoke-data. Features: 220m LLM, VRAM: 0.4GB, Context: 2K, License: apache-2.0, LLM Explorer Score: 0.17, Arc: 24.8, HellaSwag: 29.8, MMLU: 25.9, GSM8K: 0.7.

Dataset:bee-spoke-data/knowled... Dataset:jeankaddour/minipile Dataset:mattymchen/refinedweb-... Dataset:pszemraj/simple wikipe... En Endpoints compatible Llama Llama2 Model-index Region:us Safetensors Smol llama

Model Card on HF 🤗: https://huggingface.co/BEE-spoke-data/smol_llama-220M-GQA

Smol Llama 220M GQA Parameters and Internals

Model Type

decoder

Additional Notes

This is the first version of a small 220M parameter model. Fine-tuning options are available for further customization.

Training Details

Data Sources:

JeanKaddour/minipile, pszemraj/simple_wikipedia_LM, mattymchen/refinedweb-3m, BEE-spoke-data/knowledge-inoc-concat-v1

Context Length:

2048

Hardware Used:

one GPU

Model Architecture:

1024 hidden size, 10 layers, GQA (32 heads, 8 key-value)

LLM Name	Smol Llama 220M GQA
Repository 🤗	https://huggingface.co/BEE-spoke-data/smol_llama-220M-GQA
Model Size	220m
Required VRAM	0.4 GB
Updated	2026-08-13
Maintainer	BEE-spoke-data
Model Type	llama
Model Files	0.4 GB
Supported Languages	en
Model Architecture	LlamaForCausalLM
License	apache-2.0
Context Length	2048
Model Max Length	2048
Transformers Version	4.37.0.dev0
Tokenizer Class	LlamaTokenizer
Padding Token	</s>
Vocabulary Size	32128
Torch Data Type	bfloat16

Rank the Smol Llama 220M GQA 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 Smol Llama 220M GQA

Best Alternatives	Context / RAM	Downloads	Likes
...l Llama 220M GQA 32K Theta Sft	32K / 0.4 GB	6	2
Smol Llama 220M GQA 32K Theta	32K / 0.4 GB	8	1
Nuevamol 220M Base	8K / 0.4 GB	6	0
Saily 220B	4K / 417 GB	96	20
Smol Llama 220M Openhermes	2K / 0.4 GB	113	5
Smol Llama 220M Open Instruct	2K / 0.4 GB	6	2
...mol Llama 220M GQA Fineweb Edu	2K / 0.4 GB	14	1
Smol Llama 220M Bees Internal	2K / 0.4 GB	6	1
Beecoder 220M Python	2K / 0.4 GB	12	3
Saily 220B GPTQ	4K / 105.2 GB	4	1

Note: green Score (e.g. "73.2") means that the model is better than BEE-spoke-data/smol_llama-220M-GQA.