Viking 7B By LumiOpen: Benchmarks, Features and Detailed Analysis. Insights on Viking 7B.

Autotrain compatible Da Dataset:bigcode/starcoderdata Dataset:cerebras/slimpajama-62... Dataset:mc4 En Endpoints compatible Fi Is Llama Nn No Region:us Safetensors Sharded Sv Tensorflow

Model Card on HF 🤗: https://huggingface.co/LumiOpen/Viking-7B

Viking 7B Benchmarks

LLME Score: 0.20408

^nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").

What is the LLM Explorer Rank (Score)

Viking 7B Parameters and Internals

Model Type

decoder-only, transformer

Use Cases

Areas:

research, commercial applications

Limitations:

no meaningful proficiency in languages outside of English, Finnish, Swedish, Norwegian, Danish, Icelandic, and code

Considerations:

The model is partially trained, and ethical considerations should guide its application.

Supported Languages

fi (fluent), en (fluent), da (fluent), sv (fluent), no (fluent), nn (fluent), is (fluent)

Training Details

Data Sources:

cerebras/SlimPajama-627B, bigcode/starcoderdata, mc4

Data Volume:

2 trillion tokens

Methodology:

uses a LLaMA-like GPT architecture, rotary positional embeddings, flash attention

Context Length:

4096

Hardware Used:

256 AMD MI250X GPUs

Model Architecture:

decoder-only transformer

Safety Evaluation

Ethical Considerations:

Viking may produce outputs that can be considered inaccurate, prejudiced, or controversial. Users should exercise discretion.

Responsible Ai Considerations

Mitigation Strategies:

Special care should be taken when using outputs.

Input Output

Accepted Modalities:

text

Release Notes

Notes:

Training checkpoints available as branches in the repository, e.g., 100B, 200B, up to 2000B tokens.

LLM Name	Viking 7B
Repository 🤗	https://huggingface.co/LumiOpen/Viking-7B
Model Size	7b
Required VRAM	15.1 GB
Updated	2025-02-05
Maintainer	LumiOpen
Model Type	llama
Model Files	4.9 GB: 1-of-4 5.0 GB: 2-of-4 4.1 GB: 3-of-4 1.1 GB: 4-of-4
Supported Languages	fi en da sv no nn is
Model Architecture	LlamaForCausalLM
License	apache-2.0
Context Length	4096
Model Max Length	4096
Transformers Version	4.40.0
Tokenizer Class	BloomTokenizer
Padding Token	<pad>
Vocabulary Size	131072
Torch Data Type	bfloat16

Quantized Models of the Viking 7B

Model	Likes	Downloads	VRAM
...sh Alpaca Tiny V2 7B EXL2 4bpw	0	9	4 GB
Viking 7B EXL2 4bpw	1	12	4 GB

Best Alternatives to Viking 7B

Best Alternatives	Context / RAM	Downloads	Likes
...1M 1000000ctx AEZAKMI 3 1 1702	1024K / 13.5 GB	53	1
... Qwen2.5llamaify 7B V23.1 200K	195K / 15.2 GB	4950	2
LlamaStock 8B	128K / 16.1 GB	15	1
SuperNeuralDreadDevil 8B	128K / 16.1 GB	31	1
Yarn Llama 2 7B 128K	128K / 13.5 GB	5705	39
LLaMA 7B PoSE YaRN 128K	128K / 13.5 GB	8	3
LLaMA 7B PoSE Linear 96K	96K / 27 GB	8	2
LLaMA 7B PoSE YaRN 96K	96K / 13.5 GB	8	1
Chat Llama2 7B 80K	80K / 13.8 GB	6	0
Llama2 7B 80K	80K / 13.8 GB	9	0

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

Rank the Viking 7B Capabilities

🆘 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! 🌟

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? 42577 in total.

Email us: info@extractum.io. Our Privacy Policy | Terms and Conditions | Suggest an improvement.

Our Social Media →

Original data from HuggingFace, OpenCompass and various public git repos.

Release v20241227

Support LLM Explorer