Stable LM 2 1.6B
Stable LM 2 1.6B is an earlier compact base language model. It was pretrained on two trillion tokens of multilingual text and code for two epochs, before any application-specific instruction tuning.
Model design and access
The model uses partial rotary position embeddings and LayerNorm. Its training mixture includes web text, multilingual corpora and code. The card recommends fine-tuning the base for downstream tasks; it is not the same checkpoint as the Chat or Zephyr releases.Stable LM 2 1.6B official model card
Practical use
Practical guidance. This model is useful when studying a small base or adapting a narrow text task locally. Its short context calls for explicit document selection and chunking. Evaluate the adapted behavior on the required language and format instead of assuming that multilingual pretraining creates a reliable general assistant.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | stablelm-2-1_6b |
| Context window | 4,096 tokens Published checkpoint sequence length; prompt and generated text share the window. Stable LM 2 1.6B official model card |
| Parameters | 1,644,417,024.Stable LM 2 1.6B official model card |
| Architecture | 24 decoder layers; Arcade100k tokenizer.Stable LM 2 1.6B official model card |
| License | Stability AI Community License.Stable LM 2 1.6B official model card |
Related models
- Arabic Stable LM 2 Base 1.6B
Stability AI · Stable LM
Arabic Stable LM 2 Base 1.6B adapts Stable LM 2 1.6B with more than 100 billion tokens of Arabic text.
Launch date not confirmedResearch releaseDownloadable - Japanese Stable LM 2 Base 1.6B
Stability AI · Stable LM
Japanese Stable LM 2 Base 1.6B continues training the multilingual Stable LM 2 base on Japanese data.
Launch date not confirmedResearch releaseDownloadable - Stable LM 2 Chat 1.6B
Stability AI · Stable LM
Stable LM 2 Chat 1.6B is an English instruction-tuned version of the 1.6B base.
Launch date not confirmedResearch releaseDownloadable