{"data":{"slug":"stability-ai-stablelm","name":"StableLM","tagline":"Language models for development and research","github_url":"https://github.com/Stability-AI/StableLM","owner":"Stability-AI","repo":"StableLM","owner_avatar_url":"https://avatars.githubusercontent.com/u/100950301?v=4","primary_language":"Jupyter Notebook","stars":15684,"forks":1001,"topics":[],"archived":false,"github_pushed_at":"2024-04-08T19:26:02+00:00","maintenance_label":"Dormant","url":"https://www.graphcanon.com/tools/stability-ai-stablelm","markdown_url":"https://www.graphcanon.com/tools/stability-ai-stablelm.md","api_url":"https://www.graphcanon.com/api/graphcanon/tools/stability-ai-stablelm","graph_url":"https://www.graphcanon.com/api/graphcanon/graph?tool=stability-ai-stablelm","description":"StableLM: Stability AI Language Models","homepage_url":null,"license":"Apache-2.0","open_issues":28,"watchers":191,"ai_summary":"Repo contains Stability AI's StableLM series of language models with details on model releases and technical documentation.","readme_excerpt":"# StableLM: Stability AI Language Models\n\n\n<br/>*“A Stochastic Parrot, flat design, vector art” — [Stable Diffusion XL](https://clipdrop.co/stable-diffusion)*\n\nThis repository contains Stability AI's ongoing development of the StableLM series of language models and will be continuously updated with new checkpoints. The following provides an overview of all currently available models. More coming soon.\n\n## News\n\n*September 29, 2023*\n\n- Released StableLM-3B-4E1T model under [CC BY-SA-4.0](https://creativecommons.org/licenses/by-sa/4.0/).\n\n*August 5, 2023*\n\n- Released patched StableLM-Alpha v2 models with 3B and 7B parameters.\n\n*April 28, 2023*\n\n- Released StableVicuna-13B, our RLHF fine-tune of [Vicuna-13B v0](https://huggingface.co/lmsys/vicuna-13b-delta-v0), which itself is a fine-tune of [LLaMA-13B](https://github.com/facebookresearch/llama). Delta weights over the original Llama model is released under ([CC BY-NC-SA-4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/)).\n\n*April 20, 2023*\n\n- Released initial set of StableLM-Alpha models, with 3B and 7B parameters. Base models are released under [CC BY-SA-4.0](https://creativecommons.org/licenses/by-sa/4.0/).\n\n- Try to chat with our 7B model, `StableLM-Tuned-Alpha-7B`, on [Hugging Face Spaces](https://huggingface.co/spaces/stabilityai/stablelm-tuned-alpha-chat).\n\n## Models\n\n### StableLM-3B-4E1T\n\n> Technical Report: [StableLM-3B-4E1T](https://stability.wandb.io/stability-llm/stable-lm/reports/StableLM-3B-4E1T--VmlldzoyMjU4?accessToken=u3zujipenkx5g7rtcj9qojjgxpconyjktjkli2po09nffrffdhhchq045vp0wyfo)\n\nStableLM-3B-4E1T is a 3 billion (3B) parameter language model pre-trained under the multi-epoch regime to study the impact of repeated tokens on downstream performance. Given prior success in this area ([Tay et al., 2023](https://arxiv.org/pdf/2205.05131.pdf) and [Taylor et al., 2022](https://galactica.org/static/paper.pdf)), we train on 1 trillion (1T) tokens for 4 epochs following the observations of [Muennighoff et al. (2023)](https://arxiv.org/abs/2305.16264) in \"Scaling Data-Constrained Language Models\" in which they find \"training with up to 4 epochs of repeated data yields negligible changes to loss compared to having unique data.\" Further inspiration for the token count is taken from \"Go smol or go home\" ([De Vries, 2023](https://www.harmdevries.com/post/model-size-vs-compute-overhead/)), which suggests a 2.96B model trained for 2.85 trillion tokens achieves a similar loss to a Chinchilla compute-optimal 9.87B language model ($k_n = 0.3$).\n\n| Size | StableLM-3B-4E1T                                                     | Training Tokens | Parameters    |\n|------|--------------------------------------------------------------------|-----------------|---------------|\n| 3B   | [checkpoint](https://huggingface.co/stabilityai/stablelm-3b-4e1t) | 4T              | 2,795,443,200 |\n\n#### Model Architecture\n\nThe model is a decoder-only transformer similar to the LLaMA ([Touvron et al., 2023](https://arxiv.org/abs/2307.09288)) architecture with the following modifications:\n\n| Parameters     | Hidden Size | Layers | Heads | Sequence Length |\n|----------------|-------------|--------|-------|-----------------|\n| 2,795,443,200  | 2560        | 32     | 32    | 4096            |\n\n- **Position Embeddings**: Rotary Position Embeddings ([Su et al., 2021](https://arxiv.org/abs/2104.09864)) applied to the first 25% of head embedding dimensions for improved throughput following [Black et al. (2022)](https://arxiv.org/pdf/2204.06745.pdf).\n- **Normalization**: LayerNorm ([Ba et al., 2016](https://arxiv.org/abs/1607.06450)) with learned bias terms as opposed to RMSNorm ([Zhang & Sennrich, 2019](https://arxiv.org/abs/1910.07467)).\n- **Tokenizer**: GPT-NeoX ([Black et al., 2022](https://arxiv.org/abs/2204.06745)).\n\n#### Training Data\n\nThe dataset is comprised of a filtered mixture of open-source large-scale datasets available on the [HuggingFace Hub](https://huggingface.co/datasets): Falcon Refined","github_created_at":"2023-04-19T14:02:16+00:00","created_at":"2026-07-11T23:09:10.076825+00:00","updated_at":"2026-08-01T18:00:38.532079+00:00","categories":[{"slug":"llm-frameworks","name":"LLM Frameworks","url":"https://www.graphcanon.com/categories/llm-frameworks","markdown_url":"https://www.graphcanon.com/categories/llm-frameworks.md","api_url":"https://www.graphcanon.com/api/graphcanon/categories/llm-frameworks"}],"tags":[{"slug":"ai-research","name":"ai-research"},{"slug":"language-models","name":"language-models"},{"slug":"model-training","name":"model-training"},{"slug":"open-source","name":"open-source"}],"trust":{"provenance":{"is_fork":false,"github_id":630023030,"owner_type":"Organization","methodology":"github_public_v1","parent_repo":null,"near_duplicate_slugs":[]},"computed_at":"2026-08-01T18:00:37.804Z","maintenance":{"label":"Dormant","score":18,"methodology":"github_public_v1","releases_90d":0,"days_since_push":844,"last_release_at":null},"security_summary":{"status":"no_lockfile","scanner":null,"low_count":0,"high_count":0,"last_scan_at":"2026-07-11T23:09:15.715Z","medium_count":0,"scan_profile":"none","critical_count":0}},"capability_facts":{"scan":{"source":"repo_scan","observed_at":"2026-08-01T18:00:38.259Z"},"languages":{"value":["jupyter notebook"],"source":"github.language","observed_at":"2026-08-01T18:00:38.259Z"},"license_spdx":{"value":"Apache-2.0","source":"github.license","observed_at":"2026-08-01T18:00:38.259Z"}},"decision_facts":{"hosting":null,"pricing":null,"requirements":null,"constraints":null,"when_to_use":["When targeting research into the impact of multi-epoch token repetition on model performance, as StableLM is specifically designed around this concept.","In scenarios requiring creative or generative text tasks where a 3 billion parameter model suffices for the use case while offering flexibility and custom development."],"when_not_to_use":["If your project requires the strictest data privacy guarantees since some models are under less restrictive licenses, limiting their usage in projects with such constraints.","For applications needing larger language models than 13 billion parameters, as StableLM's largest model is StableVicuna-13B."],"source":"enrich:decision_facts","observed_at":"2026-07-14T18:33:58.379Z"},"constraint_facets":null,"decision_summary":[{"label":"Adopt for","value":"StableLM offers pre-trained language models for development and research with an emphasis on repeated-token training effects to improve performance."}]}}