Swallow 7B Hf by tokyotech-llm

 ยป  All LLMs  ยป  tokyotech-llm  ยป  Swallow 7B Hf   URL Share it on

  Arxiv:2404.17733   Arxiv:2404.17790   Autotrain compatible   En   Endpoints compatible   Ja   Llama   Pytorch   Region:us   Safetensors   Sharded   Tensorflow

Swallow 7B Hf Benchmarks

nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").
Swallow 7B Hf (tokyotech-llm/Swallow-7b-hf)

Swallow 7B Hf Parameters and Internals

Model Type 
text generation
Use Cases 
Areas:
Research, Commercial applications
Limitations:
Models not tuned to ensure outputs align with human intent and safety considerations.
Additional Notes 
The models are continually pre-trained and instruction-tuned, emphasizing Japanese language capabilities.
Supported Languages 
supported_languages_list (Japanese, English), languages_details (The Swallow model has undergone continual pre-training with the addition of Japanese language data.)
Training Details 
Data Sources:
Japanese Wikipedia, RefinedWeb, Swallow Corpus, The Pile
Methodology:
Supervised fine-tuning (SFT) and instruction tuning using Anthropic HH-RLHF, Databricks Dolly 15-k, and OpenAssistant Conversations Dataset.
Model Architecture:
Refer to LLaMA-2 technical report for details on the model architecture.
Input Output 
Accepted Modalities:
text
Output Format:
Text
Performance Tips:
Model employs a tokenizer with a broadened vocabulary based on Japanese data, offering efficient text representation and faster inference.
Release Notes 
Version:
0.1
Date:
April 26, 2024
Notes:
Release of Swallow-7b-instruct-v0.1, Swallow-13b-instruct-v0.1, and Swallow-70b-instruct-v0.1.
Version:
N/A
Date:
March 2, 2024
Notes:
Release of Swallow-7b-plus-hf with twice as many Japanese tokens as Swallow-7b-hf.
Version:
N/A
Date:
February 4, 2024
Notes:
Release of Swallow-13b-NVE-hf.
Version:
N/A
Date:
January 26, 2024
Notes:
Release of Swallow-7b-NVE-hf, Swallow-7b-NVE-instruct-hf, Swallow-70b-NVE-hf, Swallow-70b-NVE-instruct-hf.
Version:
N/A
Date:
December 19, 2023
Notes:
Release of Swallow-7b-hf, Swallow-7b-instruct-hf, Swallow-13b-hf, Swallow-13b-instruct-hf, Swallow-70b-hf, Swallow-70b-instruct-hf.
LLM NameSwallow 7B Hf
Repository ๐Ÿค—https://huggingface.co/tokyotech-llm/Swallow-7b-hf 
Model Size7b
Required VRAM13.7 GB
Updated2025-02-09
Maintainertokyotech-llm
Model Typellama
Model Files  10.0 GB: 1-of-2   3.7 GB: 2-of-2   10.0 GB: 1-of-2   3.7 GB: 2-of-2
Supported Languagesen ja
Model ArchitectureLlamaForCausalLM
Licensellama2
Context Length4096
Model Max Length4096
Transformers Version4.33.2
Tokenizer ClassLlamaTokenizer
Beginning of Sentence Token<s>
End of Sentence Token</s>
Unk Token<unk>
Vocabulary Size43176
Torch Data Typebfloat16

Quantized Models of the Swallow 7B Hf

Model
Likes
Downloads
VRAM
Swallow 7B GGUF42442 GB
Swallow 7B GPTQ194 GB
Swallow 7B AWQ0104 GB

Best Alternatives to Swallow 7B Hf

Best Alternatives
Context / RAM
Downloads
Likes
...1M 1000000ctx AEZAKMI 3 1 17021024K / 13.5 GB481
... Qwen2.5llamaify 7B V23.1 200K195K / 15.2 GB50552
LlamaStock 8B128K / 16.1 GB101
SuperNeuralDreadDevil 8B128K / 16.1 GB431
Yarn Llama 2 7B 128K128K / 13.5 GB590939
LLaMA 7B PoSE YaRN 128K128K / 13.5 GB83
LLaMA 7B PoSE Linear 96K96K / 27 GB82
LLaMA 7B PoSE YaRN 96K96K / 13.5 GB101
Chat Llama2 7B 80K80K / 13.8 GB90
Llama2 7B 80K80K / 13.8 GB90
Note: green Score (e.g. "73.2") means that the model is better than tokyotech-llm/Swallow-7b-hf.

Rank the Swallow 7B Hf Capabilities

๐Ÿ†˜ Have you tried this model? Rate its performance. This feedback would greatly assist ML community in identifying the most suitable model for their needs. Your contribution really does make a difference! ๐ŸŒŸ

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  

What open-source LLMs or SLMs are you in search of? 42848 in total.

Our Social Media →  
Original data from HuggingFace, OpenCompass and various public git repos.
Release v20241227