What are the hardware requirements for Smol Llama 101M GQA?

Smol Llama 101M GQA requires approximately 0.4 GB of VRAM and supports a context window of 1K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Smol Llama 101M GQA and how large is it?

Smol Llama 101M GQA is developed by BEE-spoke-data, a model with 101m parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Smol Llama 101M GQA perform on standard benchmarks?

Smol Llama 101M GQA has the following published scores: MMLU 24.35. Compare against reference models on this page or on the LLM Explorer leaderboards.

Smol Llama 101M GQA by BEE-spoke-data — VRAM 0.4GB, 1K context

Name: Smol Llama 101M GQA
Author: BEE-spoke-data

Smol Llama 101M GQA is an open-source language model by BEE-spoke-data. Features: 101m LLM, VRAM: 0.4GB, Context: 1K, License: apache-2.0, LLM Explorer Score: 0.16, Arc: 23.5, HellaSwag: 28.7, MMLU: 24.4, GSM8K: 0.8.

Dataset:bee-spoke-data/wikiped... Dataset:jeankaddour/minipile Dataset:mattymchen/refinedweb-... Dataset:pszemraj/simple wikipe... Doi:10.57967/hf/1440 En Endpoints compatible Llama Llama2 Region:us Safetensors Smol llama

Model Card on HF 🤗: https://huggingface.co/BEE-spoke-data/smol_llama-101M-GQA

Smol Llama 101M GQA Parameters and Internals

Model Type

text-generation

Use Cases

Considerations:

Fine-tuning is recommended for specific tasks.

Additional Notes

The checkpoint is the 'raw' pre-trained model and has not been tuned to a more specific task, indicating it should be fine-tuned before use in most cases.

Supported Languages

en (English)

Training Details

Data Sources:

JeanKaddour/minipile, pszemraj/simple_wikipedia_LM, BEE-spoke-data/wikipedia-20230901.en-deduped, mattymchen/refinedweb-3m

Methodology:

train-from-scratch

Context Length:

1024

Training Time:

5 compute-days

Hardware Used:

one GPU

Model Architecture:

768 hidden size, 6 layers, GQA (24 heads, 8 key-value)

Release Notes

Version:

Revision 9c9c090

Date:

2023

Notes:

smol_llama-101M-GQA (First version), a small 101M param decoder model.

LLM Name	Smol Llama 101M GQA
Repository 🤗	https://huggingface.co/BEE-spoke-data/smol_llama-101M-GQA
Model Size	101m
Required VRAM	0.4 GB
Updated	2026-08-24
Maintainer	BEE-spoke-data
Model Type	llama
Model Files	0.4 GB
Supported Languages	en
Model Architecture	LlamaForCausalLM
License	apache-2.0
Context Length	1024
Model Max Length	1024
Transformers Version	4.34.1
Tokenizer Class	LlamaTokenizer
Vocabulary Size	32128
Torch Data Type	float32

Rank the Smol Llama 101M 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 101M GQA

Best Alternatives	Context / RAM	Downloads	Likes
Smol Llama 101M Chat V1	1K / 0.4 GB	85	9
...ephyr Smol Llama 100M Sft Full	1K / 0.2 GB	97	1
...Llama 101M Midjourney Messages	1K / 0.4 GB	13	1
Smol Llama 101M GQA Python	1K / 0.4 GB	14	1