Name: StableBeluga2
Author: stabilityai

StableBeluga2 is an open-source language model by stabilityai. Features: LLM, VRAM: 275.3GB, Context: 4K, HF Score: 67.4, LLM Explorer Score: 0.21, Arc: 91.2, HellaSwag: 80.5, MMLU: 68.6, TruthfulQA: 59.4, WinoGrande: 83, GSM8K: 69.6.

Arxiv:2306.02707 Arxiv:2307.09288 Dataset:conceptofmind/cot subm... Dataset:conceptofmind/flan2021... Dataset:conceptofmind/niv2 sub... Dataset:conceptofmind/t0 submi... En Endpoints compatible Llama Pytorch Region:us Sharded

Model Card on HF 🤗: https://huggingface.co/stabilityai/StableBeluga2

StableBeluga2 Benchmarks

IFEval: 37.87 vs 88 (so35)^-57%

ARC: 91.2 vs 96.7 (so35)^-5.7%

HellaSwag: 80.5 vs 95.3 (gpt4)^-15.5%

MMLU: 68.6 vs 88.3 (so35)^-22.3%

TruthfulQA: 59.44 vs 59 (gpt4)^0.7%

WinoGrande: 82.95 vs 87.5 (gpt4)^-5.2%

GSM8K: 69.6 vs 96.4 (so35)^-27.8%

OpenCompas: 60.9 vs 73.3 (gpt4)^-16.9%

Exam: 71.9 vs 77.2 (gpt4)^-6.9%

Language: 44 vs 62 (gpt4)^-29%

Knowledge: 62.2 vs 73.5 (gpt4)^-15.4%

Understand: 72.5 vs 70 (gpt4)^3.6%

Reasoning: 58 vs 74.4 (gpt4)^-22%

MATH Lvl 5: 4.38

LLME Score: 0.21385

^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)

StableBeluga2 (stabilityai/StableBeluga2)

🌟 Advertise your project 🚀

StableBeluga2 Parameters and Internals

Model Type

text generation, auto-regressive language model

Use Cases

Areas:

research, commercial applications

Limitations:

Potential for producing inaccurate, biased, or objectionable responses.

Considerations:

Testing conducted in English, developers should perform additional safety testing before deployment.

Supported Languages

English (full)

Training Details

Data Sources:

conceptofmind/cot_submix_original, conceptofmind/flan2021_submix_original, conceptofmind/t0_submix_original, conceptofmind/niv2_submix_original

Methodology:

Supervised fine-tuning on an internal Orca-style dataset.

Model Architecture:

Fine-tuned Llama2 70B

Responsible Ai Considerations

Fairness:

Beluga's potential outputs cannot be predicted in advance, and the model may produce inaccurate, biased or other objectionable responses.

Mitigation Strategies:

Developers should perform safety testing and tuning tailored to their specific applications.

Input Output

Input Format:

System and User prompt

Accepted Modalities:

text

Output Format:

Assistant response

Performance Tips:

Use system prompts as recommended.

LLM Name	StableBeluga2
Repository 🤗	https://huggingface.co/stabilityai/StableBeluga2
Required VRAM	275.3 GB
Updated	2026-04-02
Maintainer	stabilityai
Model Type	llama
Model Files	9.4 GB: 1-of-29 9.3 GB: 2-of-29 10.0 GB: 3-of-29 9.9 GB: 4-of-29 9.3 GB: 5-of-29 9.3 GB: 6-of-29 9.3 GB: 7-of-29 10.0 GB: 8-of-29 9.9 GB: 9-of-29 9.3 GB: 10-of-29 9.3 GB: 11-of-29 9.3 GB: 12-of-29 10.0 GB: 13-of-29 9.9 GB: 14-of-29 9.3 GB: 15-of-29 9.3 GB: 16-of-29 9.3 GB: 17-of-29 10.0 GB: 18-of-29 9.9 GB: 19-of-29 9.3 GB: 20-of-29 9.3 GB: 21-of-29 9.3 GB: 22-of-29 10.0 GB: 23-of-29 9.9 GB: 24-of-29 9.3 GB: 25-of-29 9.3 GB: 26-of-29 9.3 GB: 27-of-29 10.0 GB: 28-of-29 7.6 GB: 29-of-29
Supported Languages	en
Model Architecture	LlamaForCausalLM
Context Length	4096
Model Max Length	4096
Transformers Version	4.32.0.dev0
Tokenizer Class	LlamaTokenizer
Beginning of Sentence Token	<s>
End of Sentence Token	</s>
Unk Token	<unk>
Vocabulary Size	32000
Torch Data Type	float32

Quantized Models of the StableBeluga2

Model	Likes	Downloads	VRAM
StableBeluga2 70B AWQ	2	5	36 GB
StableBeluga2 70B GGUF	4	81	29 GB
StableBeluga2 70B GPTQ	90	129	35 GB
StableBeluga2 70B GGML	40	20	28 GB

Best Alternatives to StableBeluga2

Best Alternatives	Context / RAM	Downloads	Likes
LWM Text Chat 512K	512K / 13.5 GB	7	2
LWM Text 512K	512K / 13.5 GB	5	2
LWM Text 256K	256K / 13.5 GB	11	3
LWM Text Chat 256K	256K / 13.5 GB	6	3
Pallas 0.5 LASER 0.1	195K / 68.9 GB	1888	2
Orpheus AUS	128K / GB	5	0
Llama 3.1 Medprm Reward V1.0	128K / 16.1 GB	94	16
Finetuning Health Ci	128K / 6.5 GB	2	2
Ukr Synth Phi 3.5	128K / 7.6 GB	7	0
WizardLM Phi 3.5	128K / 7.6 GB	5	0

Rank the StableBeluga2 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? 52473 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.

Check out Ag3ntum — our secure, self-hosted AI agent for server management.

Release v20260328a

Support LLM Explorer

StableBeluga2 by stabilityai

» All LLMs » stabilityai » StableBeluga2 URL Share it on