What are the hardware requirements for Smaug 72B V0.1?

Smaug 72B V0.1 requires approximately 144.5 GB of VRAM and supports a context window of 32K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Smaug 72B V0.1 and how large is it?

Smaug 72B V0.1 is developed by abacusai, a model with 72b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Smaug 72B V0.1 perform on standard benchmarks?

Smaug 72B V0.1 has the following published scores: MMLU 77.15. Compare against reference models on this page or on the LLM Explorer leaderboards.

Smaug 72B V0.1 by abacusai — VRAM 144.5GB, 32K context

Name: Smaug 72B V0.1
Author: abacusai

Smaug 72B V0.1 is an open-source language model by abacusai. Features: 72b LLM, VRAM: 144.5GB, Context: 32K, License: other, LLM Explorer Score: 0.24, Arc: 76, HellaSwag: 89.3, MMLU: 77.2, GSM8K: 78.7.

Arxiv:2402.13228 Base model:finetune:moreh/momo... Base model:moreh/momo-72b-lora... Endpoints compatible Llama Lora Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/abacusai/Smaug-72B-v0.1

Smaug 72B V0.1 Parameters and Internals

Additional Notes

Utilizes DPOP, a new fine-tuning technique that outperforms standard DPO across several datasets and tasks.

Training Details

Methodology:

DPOP - a new fine-tuning technique

LLM Name	Smaug 72B V0.1
Repository 🤗	https://huggingface.co/abacusai/Smaug-72B-v0.1
Base Model(s)	moreh/MoMo-72B-lora-1.8.7-DPO moreh/MoMo-72B-lora-1.8.7-DPO
Model Size	72b
Required VRAM	144.5 GB
Updated	2026-07-26
Maintainer	abacusai
Model Type	llama
Model Files	4.8 GB: 1-of-30 5.0 GB: 2-of-30 5.0 GB: 3-of-30 4.8 GB: 4-of-30 4.8 GB: 5-of-30 4.8 GB: 6-of-30 5.0 GB: 7-of-30 5.0 GB: 8-of-30 4.8 GB: 9-of-30 4.8 GB: 10-of-30 4.8 GB: 11-of-30 5.0 GB: 12-of-30 5.0 GB: 13-of-30 4.8 GB: 14-of-30 4.8 GB: 15-of-30 4.8 GB: 16-of-30 5.0 GB: 17-of-30 5.0 GB: 18-of-30 4.8 GB: 19-of-30 4.8 GB: 20-of-30 4.8 GB: 21-of-30 5.0 GB: 22-of-30 5.0 GB: 23-of-30 4.8 GB: 24-of-30 4.8 GB: 25-of-30 4.8 GB: 26-of-30 5.0 GB: 27-of-30 5.0 GB: 28-of-30 4.8 GB: 29-of-30 2.9 GB: 30-of-30
Model Architecture	LlamaForCausalLM
License	other
Context Length	32768
Model Max Length	32768
Transformers Version	4.36.2
Tokenizer Class	GPT2Tokenizer
Vocabulary Size	152064
LoRA Model	Yes
Torch Data Type	bfloat16

Rank the Smaug 72B V0.1 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 Smaug 72B V0.1

Model	Likes	Downloads	VRAM
Smaug 72B V0.1 AWQ	3	4	41 GB
Smaug 72B V0.1 GPTQ	2	5	41 GB

Best Alternatives to Smaug 72B V0.1

Best Alternatives	Context / RAM	Downloads	Likes
2 Pro Math	128K / 141.9 GB	10	0
TW3 JRGL V2	32K / 79.7 GB	16209	0
Le Triomphant ECE TW3	32K / 79.7 GB	16266	5
ECE TW3 JRGL V5	32K / 159.6 GB	8245	1
MoMo 72B Lora 1.8.7 DPO	32K / 208.5 GB	314	68
Rhea 72B V0.5	32K / 144.5 GB	8194	134
JuliusCesar 72B BeyonderV.0	32K / 74.2 GB	77	0
Caigun Model 72B KGI	32K / 144.6 GB	31	0
Rhea 125 V0.5	32K / 249 GB	13	1
Exodius 70B	32K / 144.6 GB	8	0

Note: green Score (e.g. "73.2") means that the model is better than abacusai/Smaug-72B-v0.1.