What are the hardware requirements for Stablelm 2 12B?

Stablelm 2 12B requires approximately 24.4 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 Stablelm 2 12B and how large is it?

Stablelm 2 12B is developed by stabilityai, a model with 12b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Stablelm 2 12B perform on standard benchmarks?

Stablelm 2 12B has the following published scores: MMLU 62.04. Compare against reference models on this page or on the LLM Explorer leaderboards.

Stablelm 2 12B by stabilityai — VRAM 24.4GB, 4K context

Name: Stablelm 2 12B
Author: stabilityai

Stablelm 2 12B is an open-source language model by stabilityai. Features: 12b LLM, VRAM: 24.4GB, Context: 4K, License: other, LLM Explorer Score: 0.2, Arc: 58.5, HellaSwag: 84.3, MMLU: 62, GSM8K: 56.

Arxiv:1607.06450 Arxiv:1910.02054 Arxiv:2101.00027 Arxiv:2104.09864 Arxiv:2204.06745 Arxiv:2206.11147 Arxiv:2302.05442 Arxiv:2305.06161 Arxiv:2305.14201 Arxiv:2309.09400 Arxiv:2309.14322 Arxiv:2402.17834 Dataset:bigcode/starcoderdata Dataset:carperai/pilev2-dev Dataset:dataprovenanceinitiati... Dataset:tiiuae/falcon-refinedw... Dataset:togethercomputer/redpa... Dataset:uonlp/culturax De En Endpoints compatible Es Fr It Nl Pt Region:us Safetensors Sharded Stablelm Tensorflow

Model Card on HF 🤗: https://huggingface.co/stabilityai/stablelm-2-12b

Stablelm 2 12B Parameters and Internals

Model Type

causal-lm

Use Cases

Areas:

research, commercial applications

Limitations:

unreliable outputs, unsafe behaviors, offensive content

Considerations:

The model may exhibit undesirable behaviors and should be fine-tuned and evaluated for specific applications.

Additional Notes

For questions and comments about the model, email: lm@stability.ai

Supported Languages

en (fluent), de (fluent), es (fluent), fr (fluent), it (fluent), nl (fluent), pt (fluent)

Training Details

Data Sources:

tiiuae/falcon-refinedweb, togethercomputer/RedPajama-Data-1T, uonlp/CulturaX, CarperAI/pilev2-dev, bigcode/starcoderdata, DataProvenanceInitiative/Commercially-Verified-Licenses

Data Volume:

2 trillion tokens

Context Length:

4096

Hardware Used:

384 NVIDIA H100 GPUs (AWS P5 instances)

Model Architecture:

decoder-only transformer with 12.1 billion parameters, 5120 hidden size, 40 layers, and 32 attention heads.

Input Output

Input Format:

prompts as plain text

Accepted Modalities:

text

Output Format:

generated text from the model

LLM Name	Stablelm 2 12B
Repository 🤗	https://huggingface.co/stabilityai/stablelm-2-12b
Model Size	12b
Required VRAM	24.4 GB
Updated	2026-07-20
Maintainer	stabilityai
Model Type	stablelm
Model Files	5.0 GB: 1-of-5 5.0 GB: 2-of-5 5.0 GB: 3-of-5 5.0 GB: 4-of-5 4.4 GB: 5-of-5
Supported Languages	en de es fr it nl pt
Model Architecture	StableLmForCausalLM
License	other
Context Length	4096
Model Max Length	4096
Transformers Version	4.40.0
Tokenizer Class	GPT2TokenizerFast
Padding Token	<\|endoftext\|>
Vocabulary Size	100352
Torch Data Type	bfloat16

Rank the Stablelm 2 12B 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 Stablelm 2 12B

Best Alternatives	Context / RAM	Downloads	Likes
Stablelm 2 12B Chat	4K / 24.4 GB	215	88