What are the hardware requirements for Swallow 70B Instruct AWQ?

Swallow 70B Instruct AWQ requires approximately 37 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 Swallow 70B Instruct AWQ and how large is it?

Swallow 70B Instruct AWQ is developed by TheBloke, a model with 70b 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 Swallow 70B Instruct AWQ?

Swallow 70B Instruct AWQ is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

Swallow 70B Instruct AWQ by TheBloke — VRAM 37GB, 4K context

Name: Swallow 70B Instruct AWQ
Author: TheBloke

Swallow 70B Instruct AWQ is an open-source language model by TheBloke. Features: 70b LLM, VRAM: 37GB, Context: 4K, License: llama2, Quantized, Instruction-Based, LLM Explorer Score: 0.1.

4-bit Awq Base model:quantized:tokyotech... Base model:tokyotech-llm/swall... En Instruct Ja Llama Quantized Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/TheBloke/Swallow-70B-instruct-AWQ

Swallow 70B Instruct AWQ Parameters and Internals

Model Type

llama

Use Cases

Applications:

Research, Commercial applications

Additional Notes

The Swallow model is mainly tuned for Japanese language data and features modified vocabulary to enhance inference speed.

Supported Languages

languages_supported (>2 including Japanese and English.), proficiency_notes (>Proficient in Japanese and English.)

Training Details

Data Sources:

Japanese Wikipedia, RefinedWeb, Swallow Corpus, The Pile

Methodology:

Supervised fine-tuning (SFT)

Context Length:

4096

Model Architecture:

Llama

Input Output

Input Format:

Formatted Japanese instructions and prompts.

Accepted Modalities:

Text

Output Format:

Generated textual response

LLM Name	Swallow 70B Instruct AWQ
Repository 🤗	https://huggingface.co/TheBloke/Swallow-70B-instruct-AWQ
Model Name	Swallow 70B Instruct
Model Creator	tokyotech-llm
Base Model(s)	Swallow 70B Instruct Hf tokyotech-llm/Swallow-70b-instruct-hf
Model Size	70b
Required VRAM	37 GB
Updated	2026-07-22
Maintainer	TheBloke
Model Type	llama
Instruction-Based	Yes
Model Files	9.9 GB: 1-of-4 10.0 GB: 2-of-4 9.9 GB: 3-of-4 7.2 GB: 4-of-4
Supported Languages	en ja
AWQ Quantization	Yes
Quantization Type	awq
Model Architecture	LlamaForCausalLM
License	llama2
Context Length	4096
Model Max Length	4096
Transformers Version	4.37.0.dev0
Tokenizer Class	LlamaTokenizer
Beginning of Sentence Token	<s>
End of Sentence Token	</s>
Unk Token	<unk>
Vocabulary Size	43176
Torch Data Type	float16

Rank the Swallow 70B Instruct AWQ 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

Best Alternatives to Swallow 70B Instruct AWQ

Best Alternatives	Context / RAM	Downloads	Likes
...0B Instruct Gradient 1048K AWQ	1024K / 39.9 GB	5	1
...70B Instruct Gradient 262K AWQ	256K / 39.9 GB	5	0
Llama 3.3 70B Instruct AWQ	128K / 39.9 GB	262002	46
...lama 3.3 70B Instruct AWQ INT4	128K / 39.9 GB	82744	30
Llama 3.3 70B Instruct AWQ	128K / 39.9 GB	16090	12
... SauerkrautLM 70B Instruct AWQ	128K / 39.9 GB	5	5
Llama 3 70B Instruct AWQ	8K / 39.9 GB	122175	71
Meta Llama 3 70B Instruct AWQ	8K / 39.9 GB	8	1
Llama 3 70B Instruct AWQ	8K / 39.9 GB	17	1
... Llama 3 70B Instruct AWQ GEMM	8K / 39.9 GB	6	0

Note: green Score (e.g. "73.2") means that the model is better than TheBloke/Swallow-70B-instruct-AWQ.