What are the hardware requirements for Llama 3 Smaug 8B?

Llama 3 Smaug 8B requires approximately 16.1 GB of VRAM and supports a context window of 8K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Llama 3 Smaug 8B and how large is it?

Llama 3 Smaug 8B is developed by abacusai, a model with 8b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Llama 3 Smaug 8B perform on standard benchmarks?

Llama 3 Smaug 8B has the following published scores: MMLU 65.77. Compare against reference models on this page or on the LLM Explorer leaderboards.

Llama 3 Smaug 8B by abacusai — VRAM 16.1GB, 8K context

Name: Llama 3 Smaug 8B
Author: abacusai

Llama 3 Smaug 8B is an open-source language model by abacusai. Features: 8b LLM, VRAM: 16.1GB, Context: 8K, License: llama2, LLM Explorer Score: 0.22, Arc: 59.6, HellaSwag: 80, MMLU: 65.8, GSM8K: 55.9.

Arxiv:2402.13228 Conversational Dataset:anon8231489123/sharegp... Dataset:aqua rat Dataset:m-a-p/codefeedback-fil... Dataset:microsoft/orca-math-wo... Endpoints compatible Eval-results Llama Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/abacusai/Llama-3-Smaug-8B

Llama 3 Smaug 8B Parameters and Internals

Model Type

text generation

Additional Notes

This version of Smaug uses new techniques and new data compared to Smaug-72B, and more information will be released later.

Training Details

Data Sources:

aqua_rat, microsoft/orca-math-word-problems-200k, m-a-p/CodeFeedback-Filtered-Instruction, anon8231489123/ShareGPT_Vicuna_unfiltered

Methodology:

Smaug recipe for improving performance on real-world multi-turn conversations

LLM Name	Llama 3 Smaug 8B
Repository 🤗	https://huggingface.co/abacusai/Llama-3-Smaug-8B
Model Size	8b
Required VRAM	16.1 GB
Updated	2026-07-27
Maintainer	abacusai
Model Type	llama
Model Files	5.0 GB: 1-of-4 5.0 GB: 2-of-4 4.9 GB: 3-of-4 1.2 GB: 4-of-4
Model Architecture	LlamaForCausalLM
License	llama2
Context Length	8192
Model Max Length	8192
Transformers Version	4.37.2
Tokenizer Class	PreTrainedTokenizerFast
Padding Token	<\|end_of_text\|>
Vocabulary Size	128256
Torch Data Type	bfloat16

Rank the Llama 3 Smaug 8B 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

Quantized Models of the Llama 3 Smaug 8B

Model	Likes	Downloads	VRAM
Smaugredq	0	38	4 GB
Llama 3 Smaug 8B AWQ	1	5	5 GB
Llama 3 Smaug 8B 4bit	1	16	5 GB
Llama 3 Smaug 8B GGUF	5	535	3 GB
Smaugredq	0	0	4 GB

Best Alternatives to Llama 3 Smaug 8B

Best Alternatives	Context / RAM	Downloads	Likes
...otron 8B UltraLong 4M Instruct	4192K / 32.1 GB	1135	125
UltraLong Thinking	4192K / 16.1 GB	2	3
...a 3.1 8B UltraLong 4M Instruct	4192K / 32.1 GB	176	24
...a 3.1 8B UltraLong 2M Instruct	2096K / 32.1 GB	875	9
...otron 8B UltraLong 2M Instruct	2096K / 32.1 GB	124	18
Cthulhu 8B V1.4	1048K / 16.1 GB	10	10
...raLong 1M Instruct Abliterated	1048K / 32.1 GB	4	9
...a 3.1 8B UltraLong 1M Instruct	1048K / 32.1 GB	1387	29
...otron 8B UltraLong 1M Instruct	1048K / 32.1 GB	702	59
...xis Bookwriter Llama3.1 8B Sft	1048K / 16.1 GB	25	4