What are the hardware requirements for ALMA 7B?

ALMA 7B requires approximately 27 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 ALMA 7B and how large is it?

ALMA 7B is developed by haoranxu, a model with 7b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does ALMA 7B perform on standard benchmarks?

ALMA 7B has the following published scores: MMLU 38.04. Compare against reference models on this page or on the LLM Explorer leaderboards.

ALMA 7B by haoranxu — VRAM 27GB, 4K context

Name: ALMA 7B
Author: haoranxu

ALMA 7B is an open-source language model by haoranxu. Features: 7b LLM, VRAM: 27GB, Context: 4K, License: mit, LLM Explorer Score: 0.12, Arc: 50.3, HellaSwag: 75.5, MMLU: 38.

Arxiv:2309.11674 Arxiv:2401.08417 Endpoints compatible Llama Pytorch Region:us Sharded

Model Card on HF 🤗: https://huggingface.co/haoranxu/ALMA-7B

ALMA 7B Parameters and Internals

Model Type

Translation, LLM-based

Use Cases

Areas:

Machine Translation

Applications:

Translation services, Language learning

Additional Notes

Pretrained models should be used with LoRA models for optimal translation.

Training Details

Data Sources:

human-written parallel data, triplet preference data, monolingual data

Data Volume:

20B monolingual tokens for ALMA-7B and 12B monolingual tokens for ALMA-13B

Methodology:

Fine-tuning on monolingual data followed by parallel data optimization

Input Output

Input Format:

Text prompt with source and target languages specified

Accepted Modalities:

Text

Output Format:

Translated text

Release Notes

Version:

ALMA-7B

Date:

2023

Notes:

Fine-tuned on 20B monolingual tokens followed by parallel data optimization.

Version:

ALMA-7B-LoRA

Date:

2023

Notes:

LoRA fine-tuning on human-written parallel data.

Version:

ALMA-7B-R

Date:

2024

Notes:

Further LoRA fine-tuning with contrastive preference optimization.

Version:

ALMA-13B

Date:

2023

Notes:

Fine-tuned on 12B monolingual tokens followed by parallel data optimization.

Version:

ALMA-13B-LoRA

Date:

2023

Notes:

LoRA fine-tuning on human-written parallel data.

Version:

ALMA-13B-R

Date:

2024

Notes:

Further LoRA fine-tuning with contrastive preference optimization.

LLM Name	ALMA 7B
Repository 🤗	https://huggingface.co/haoranxu/ALMA-7B
Model Size	7b
Required VRAM	27 GB
Updated	2026-07-20
Maintainer	haoranxu
Model Type	llama
Model Files	9.9 GB: 1-of-3 9.9 GB: 2-of-3 7.2 GB: 3-of-3
Model Architecture	LlamaForCausalLM
License	mit
Context Length	4096
Model Max Length	4096
Transformers Version	4.30.0.dev0
Tokenizer Class	LlamaTokenizer
Beginning of Sentence Token	<s>
End of Sentence Token	</s>
Unk Token	<unk>
Vocabulary Size	32000
Torch Data Type	float32

Rank the ALMA 7B 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 ALMA 7B

Model	Likes	Downloads	VRAM
ALMA 7B GGUF	3	314	2 GB
ALMA 7B AWQ	1	8	3 GB
ALMA 7B GPTQ	1	5	3 GB

Best Alternatives to ALMA 7B

Best Alternatives	Context / RAM	Downloads
124	1024K / 16.1 GB	93
162	1024K / 16.1 GB	60
157	1024K / 16.1 GB	101
118	1024K / 16.1 GB	15
A5.4	1024K / 16.1 GB	12
A3.4	1024K / 16.1 GB	13
A2.4	1024K / 16.1 GB	12
A6 L	1024K / 16.1 GB	201
M	1024K / 16.1 GB	127
2 Very Sci Fi	1024K / 16.1 GB	317

Note: green Score (e.g. "73.2") means that the model is better than haoranxu/ALMA-7B.