What are the hardware requirements for Llama 2 70B Hf?

Llama 2 70B Hf requires approximately 138 GB of VRAM and supports a context window of 4K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Llama 2 70B Hf and how large is it?

Llama 2 70B Hf is developed by meta-llama, a model with 70b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Llama 2 70B Hf perform on standard benchmarks?

Llama 2 70B Hf has the following published scores: MMLU 69.83. Compare against reference models on this page or on the LLM Explorer leaderboards.

Llama 2 70B Hf by meta-llama — VRAM 138GB, 4K context

Name: Llama 2 70B Hf
Author: meta-llama

Llama 2 70B Hf is an open-source language model by meta-llama. Features: 70b LLM, VRAM: 138GB, Context: 4K, License: llama2, LLM Explorer Score: 0.19, Arc: 67.3, HellaSwag: 87.3, MMLU: 69.8, GSM8K: 54.1.

Arxiv:2307.09288 Deploy:sagemaker En Endpoints compatible Facebook Llama Llama2 Meta Pytorch Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/meta-llama/Llama-2-70b-hf

Llama 2 70B Hf Parameters and Internals

Model Type

Generative text

Use Cases

Areas:

Commercial, Research

Applications:

Assistant-like chat

Primary Use Cases:

Natural language generation tasks

Limitations:

Use cases not covered extensively in languages other than English.

Considerations:

Developers should ensure the responsible use of models.

Additional Notes

Tuned models optimized for dialogue. High carbon footprint during pretraining offset by Meta's sustainability program. Modeled potential relationships between text sequences to predict next items in sequences safely and effectively.

Supported Languages

English (Primary language for intended use)

Training Details

Data Sources:

Publicly available online data

Data Volume:

2 trillion tokens

Methodology:

Uses a mix of publicly available online data. Fine-tuned using Supervised Fine-Tuning (SFT) and Reinforcement Learning with Human Feedback (RLHF).

Context Length:

4000

Hardware Used:

A100-80GB (TDP of 350-400W)

Model Architecture:

Optimized transformer architecture

Responsible Ai Considerations

Fairness:

Model may produce inaccurate, biased, or objectionable outputs.

Transparency:

Transparency measures are in place for users.

Accountability:

Developers should perform safety testing tailored to specific applications.

Mitigation Strategies:

Safety testing and tuning recommended by Meta before deployment.

Input Output

Input Format:

Text only

Accepted Modalities:

Text

Output Format:

Text only

Performance Tips:

Specific formatting needs for chat versions, including the use of `INST` and `<>` tags, `BOS` and `EOS` tokens, and appropriate whitespace management.

LLM Name	Llama 2 70B Hf
Repository 🤗	https://huggingface.co/meta-llama/Llama-2-70b-hf
Model Size	70b
Required VRAM	138 GB
Updated	2026-07-26
Maintainer	meta-llama
Model Type	llama
Model Files	9.8 GB: 1-of-15 9.8 GB: 2-of-15 10.0 GB: 3-of-15 9.8 GB: 4-of-15 9.8 GB: 5-of-15 9.8 GB: 6-of-15 10.0 GB: 7-of-15 9.8 GB: 8-of-15 9.8 GB: 9-of-15 9.8 GB: 10-of-15 10.0 GB: 11-of-15 9.8 GB: 12-of-15 9.8 GB: 13-of-15 9.5 GB: 14-of-15 0.5 GB: 15-of-15 9.8 GB: 1-of-15 9.8 GB: 2-of-15 10.0 GB: 3-of-15 9.8 GB: 4-of-15 9.8 GB: 5-of-15 9.8 GB: 6-of-15 10.0 GB: 7-of-15 9.8 GB: 8-of-15 9.8 GB: 9-of-15 9.8 GB: 10-of-15 10.0 GB: 11-of-15 9.8 GB: 12-of-15 9.8 GB: 13-of-15 9.5 GB: 14-of-15 0.5 GB: 15-of-15
Supported Languages	en
Model Architecture	LlamaForCausalLM
License	llama2
Context Length	4096
Model Max Length	4096
Transformers Version	4.32.0.dev0
Tokenizer Class	LlamaTokenizer
Beginning of Sentence Token	<s>
End of Sentence Token	</s>
Unk Token	<unk>
Vocabulary Size	32000
Torch Data Type	float16

Rank the Llama 2 70B 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 Llama 2 70B Hf

Model	Likes	Downloads	VRAM
Llama 2 70B AWQ	14	4293	36 GB
Tulu 2 70B 6.0bpw H6 EXL2	1	3	52 GB
Tulu 2 DPO 70B 4.0bpw H6 EXL2	1	2	35 GB
Tulu 2 DPO 70B 5.0bpw H6 EXL2	2	2	43 GB
Tulu 2 DPO 70B 2.4bpw H6 EXL2	1	5	21 GB
Llama 2 70B GGUF	30	811	29 GB
Llama 2 70B GPTQ	83	485	35 GB

Best Alternatives to Llama 2 70B Hf

Best Alternatives	Context / RAM	Downloads	Likes
... Chat 1048K Chinese Llama3 70B	1024K / 141.9 GB	9069	5
... Chat 1048K Chinese Llama3 70B	1024K / 141.9 GB	7684	4
... 3 70B Instruct Gradient 1048K	1024K / 141.9 GB	22	122
Llama3 Function Calling 1048K	1024K / 141.9 GB	5	1
...a 3 70B Instruct Gradient 524K	512K / 141.9 GB	24	23
...a 3 70B Instruct Gradient 262K	256K / 141.9 GB	20	56
...ama 3 70B Arimas Story RP V2.0	256K / 141.1 GB	10	3
...ama 3 70B Arimas Story RP V1.6	256K / 141.2 GB	5	0
...ama 3 70B Arimas Story RP V1.5	256K / 141.2 GB	21	3
Yi 70B 200K RPMerge Franken	195K / 142.4 GB	10	1