Spaces:
Running
Running
| title: PresenceAI | |
| emoji: 🎭 | |
| colorFrom: purple | |
| colorTo: blue | |
| sdk: static | |
| app_file: index.html | |
| pinned: false | |
| # PresenceAI | |
| PresenceAI builds **open, production-oriented conversational models** for e-commerce, social connection, and tool-calling experiences. | |
| ## Presence Muses | |
| **Presence Muses** is our first open model family: a set of LoRA adapters covering supervised fine-tuning (SFT) and direct preference optimization (DPO). The adapters are small and composable, and are designed to sit on top of well-known base models. | |
| | Model | Base model | Stage | Adapter size | | |
| |---|---|---|---:| | |
| | [Calliope-32B-SFT-LoRA-v1.0](https://huggingface.co/PresenceAI/Calliope-32B-SFT-LoRA-v1.0) | [Qwen/Qwen3-32B](https://huggingface.co/Qwen/Qwen3-32B) | SFT | 1.07 GB | | |
| | [Polyhymnia-32B-SFT-LoRA-v1.0](https://huggingface.co/PresenceAI/Polyhymnia-32B-SFT-LoRA-v1.0) | [Qwen/Qwen3-32B](https://huggingface.co/Qwen/Qwen3-32B) | SFT (full-data run) | 1.07 GB | | |
| | [Erato-27B-DPO-LoRA-v1.0](https://huggingface.co/PresenceAI/Erato-27B-DPO-LoRA-v1.0) | [Qwen/Qwen3.8-27B](https://huggingface.co/Qwen/Qwen3.8-27B) | DPO | 638 MB | | |
| | [Thalia-31B-DPO-LoRA-v1.0](https://huggingface.co/PresenceAI/Thalia-31B-DPO-LoRA-v1.0) | [Gryphe/Gemma-4-31B-StyleTune](https://huggingface.co/Gryphe/Gemma-4-31B-StyleTune) | DPO | 935 MB | | |
| ## What we release | |
| - **LoRA adapters only** — base model weights are not redistributed; download the base model separately from its upstream repository. | |
| - **No training or evaluation datasets** are published with these adapters. | |
| - Adapters are released under **Apache-2.0**; the upstream base-model licenses and terms continue to apply. | |
| - These are research and prototyping models. Evaluate quality, safety, and policy compliance for your own deployment. | |
| ## Model lineage | |
| - **Calliope / Polyhymnia** are supervised fine-tunes of `Qwen/Qwen3-32B`; Polyhymnia is the full-data run. | |
| - **Erato** is a DPO adapter trained directly on the `Qwen/Qwen3.8-27B` base model. | |
| - **Thalia** is a DPO adapter trained on `Gryphe/Gemma-4-31B-StyleTune`, a style-tuned derivative of Gemma-4-31B — not on the original Gemma base. | |
| For prompt formats, tokenizer files, tool-calling parsers, and evaluation notes, see each model card. | |
| ## 中文简介 | |
| PresenceAI 是 Presence AI 的开源模型组织。当前发布的 **Presence Muses** 系列是四个 LoRA 适配器，覆盖 SFT 与 DPO 两个阶段，基座分别为 Qwen3-32B、Qwen3.8-27B 和 Gemma-4-31B StyleTune。我们只发布适配器权重，不发布基座和训练数据；适配器采用 Apache-2.0，基座许可继续适用。 | |
| ## Community | |
| Questions and feedback are welcome in the **Discussions** tab of the corresponding model repository. | |