What are the hardware requirements for LongAlign 13B 64K?

LongAlign 13B 64K requires approximately 26 GB of VRAM and supports a context window of 64K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed LongAlign 13B 64K and how large is it?

LongAlign 13B 64K is developed by THUDM, a model with 13b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

Where can I download or evaluate LongAlign 13B 64K?

LongAlign 13B 64K is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

LongAlign 13B 64K by THUDM — VRAM 26GB, 64K context

Name: LongAlign 13B 64K
Author: THUDM

LongAlign 13B 64K is an open-source language model by THUDM. Features: 13b LLM, VRAM: 26GB, Context: 64K, License: apache-2.0, LLM Explorer Score: 0.11.

Arxiv:2401.18058 Autotrain compatible Dataset:thudm/longalign-10k En Endpoints compatible Llama Long context Pytorch Region:us Zh

Model Card on HF 🤗: https://huggingface.co/THUDM/LongAlign-13B-64k

LongAlign 13B 64K Parameters and Internals

Model Type

text generation

Use Cases

Areas:

research

Applications:

text generation

Primary Use Cases:

instruction-following capability evaluation on queries of 10k-100k length

Additional Notes

LongAlign is the first full recipe for LLM alignment on long context. The model is open-sourced and can be used in various text generation tasks.

Supported Languages

en (High Proficiency), zh (High Proficiency)

Training Details

Data Sources:

THUDM/LongAlign-10k

Data Volume:

10,000 long instruction data of 8k-64k length

Methodology:

packing with loss weighting and sorted batching

Context Length:

64000

Input Output

Input Format:

[INST] <user_query> [/INST]

Accepted Modalities:

text

Output Format:

text

LLM Name	LongAlign 13B 64K
Repository 🤗	https://huggingface.co/THUDM/LongAlign-13B-64k
Model Size	13b
Required VRAM	26 GB
Updated	2025-07-27
Maintainer	THUDM
Model Type	llama
Model Files	26.0 GB
Supported Languages	en zh
Model Architecture	LlamaForCausalLM
License	apache-2.0
Context Length	65536
Model Max Length	65536
Transformers Version	4.33.0
Tokenizer Class	LlamaTokenizer
Beginning of Sentence Token	<s>
End of Sentence Token	</s>
Unk Token	<unk>
Vocabulary Size	32256
Torch Data Type	bfloat16

Rank the LongAlign 13B 64K 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 LongAlign 13B 64K

Model	Likes	Downloads	VRAM
LongAlign 13B 64K GGUF	3	190	4 GB
LongAlign 13B 64K AWQ	2	4	7 GB
LongAlign 13B 64K GPTQ	1	5	7 GB

Best Alternatives to LongAlign 13B 64K

Best Alternatives	Context / RAM	Downloads	Likes
Luminaura RP 13B	128K / 26 GB	9	1
Yarn Llama 2 13B 128K	128K / 26 GB	184	113
Agent Llama2 13B 80K	80K / 26.4 GB	5	0
Chat Llama2 13B 80K	80K / 52.8 GB	5	0
LongAlign 13B 64K	64K / 26 GB	96	13
LongAlign 13B 64K Base	64K / 26 GB	89	3
LongAlign 13B 64K Base	64K / 26 GB	6	3
Openbuddy Llama2 13B V15p1 64K	64K / 26.1 GB	4	4
Openbuddy Llama2 13b64k V15	64K / 26.1 GB	12	2
Yarn Llama 2 13B 64K	64K / 26 GB	99	18

Note: green Score (e.g. "73.2") means that the model is better than THUDM/LongAlign-13B-64k.