What are the hardware requirements for OpenAssistant SFT 7 Llama 30B HF?

OpenAssistant SFT 7 Llama 30B HF requires approximately 65.2 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 OpenAssistant SFT 7 Llama 30B HF and how large is it?

OpenAssistant SFT 7 Llama 30B HF is developed by TheBloke, a model with 30b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does OpenAssistant SFT 7 Llama 30B HF perform on standard benchmarks?

OpenAssistant SFT 7 Llama 30B HF has the following published scores: MMLU 57.93. Compare against reference models on this page or on the LLM Explorer leaderboards.

OpenAssistant SFT 7 Llama 30B HF by TheBloke — VRAM 65.2GB, 2K context

Name: OpenAssistant SFT 7 Llama 30B HF
Author: TheBloke

OpenAssistant SFT 7 Llama 30B HF is an open-source language model by TheBloke. Features: 30b LLM, VRAM: 65.2GB, Context: 2K, License: other, LLM Explorer Score: 0.09, Arc: 60.6, HellaSwag: 82.2, MMLU: 57.9, GSM8K: 29.8.

Arxiv:2304.07327 Endpoints compatible Llama Pytorch Region:us Sharded

Model Card on HF 🤗: https://huggingface.co/TheBloke/OpenAssistant-SFT-7-Llama-30B-HF

OpenAssistant SFT 7 Llama 30B HF Parameters and Internals

Model Type

text generation

Use Cases

Areas:

research, commercial applications

Additional Notes

This is the result of merging XORs from the repo with the original Llama 30B weights.

Supported Languages

bg (supported), ca (supported), cs (supported), da (supported), de (supported), en (supported), es (supported), fr (supported), hr (supported), hu (supported), it (supported), nl (supported), pl (supported), pt (supported), ro (supported), ru (supported), sl (supported), sr (supported), sv (supported), uk (supported)

Training Details

Data Sources:

oasst_export, vicuna, dolly15k, grade_school_math_instructions, code_alpaca

Methodology:

Merged XORs with original Llama 30B weights, using epoch 7 of OpenAssistant's training.

Context Length:

2048

Hardware Used:

unknown

Model Architecture:

llama

LLM Name	OpenAssistant SFT 7 Llama 30B HF
Repository 🤗	https://huggingface.co/TheBloke/OpenAssistant-SFT-7-Llama-30B-HF
Model Size	30b
Required VRAM	65.2 GB
Updated	2026-05-07
Maintainer	TheBloke
Model Type	llama
Model Files	9.8 GB: 1-of-7 10.0 GB: 2-of-7 9.9 GB: 3-of-7 9.9 GB: 4-of-7 9.9 GB: 5-of-7 10.0 GB: 6-of-7 5.7 GB: 7-of-7
Model Architecture	LlamaForCausalLM
License	other
Context Length	2048
Model Max Length	2048
Transformers Version	4.29.0.dev0
Tokenizer Class	LlamaTokenizer
End of Sentence Token	</s>
Unk Token	</s>
Vocabulary Size	32016
Torch Data Type	float16

Rank the OpenAssistant SFT 7 Llama 30B HF 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 OpenAssistant SFT 7 Llama 30B HF

Model	Likes	Downloads	VRAM
...Assistant SFT 7 Llama 30B GPTQ	35	124	16 GB

Best Alternatives to OpenAssistant SFT 7 Llama 30B HF

Best Alternatives	Context / RAM	Downloads	Likes
TildeOpen 30B	64K / 60.9 GB	4576	165
TildeOpen 30B 64K	64K / 60.9 GB	702	3
Tildeopen 30B Mu Instruct	64K / 60.9 GB	41	6
Flash Llama 30M 20001	32K / 0.1 GB	134	0
Smaug Slerp 30B V0.1	32K / 60.4 GB	79	0
Tenebra 30B Alpha01	16K / 65 GB	50	16
Llama33b 16K	16K / 65.2 GB	1	1
Veyra2 Mango 30M Base	4K / 0.1 GB	0	5
Yayi2 30B Llama	4K / 121.2 GB	922	22
... Tokens By Perplexity Bottom K	4K / 5.4 GB	8	0

Note: green Score (e.g. "73.2") means that the model is better than TheBloke/OpenAssistant-SFT-7-Llama-30B-HF.