{"name":"RWKV models catalog","url":"https://www.rwkv.cn/eco/models","count":26,"lastUpdated":"2025-06-20","items":[{"id":20250417,"created_at":"2025-06-20T06:28:46.76182+00:00","name":"RWKV-X","des":"基于 RWKV 的混合语言模型","category":"community","link_download":"https://huggingface.co/howard-hou/RWKV-X","link_github":"https://github.com/howard-hou/RWKV-X","link_paper":"https://arxiv.org/abs/2504.21463","link_demo":"","details":"结合 RWKV 和新型系数注意力机制，处理长上下文的能力更强，支持百万级 token 的序列编码","link_tutorial":""},{"id":20250515,"created_at":"2025-06-20T06:15:11.26781+00:00","name":"RWKV-V7-Soothe","des":"基于 RWKV-7 微调的心理咨询师模型","category":"community","link_download":"https://huggingface.co/keepzmy/RWKV-V7-Soothe","link_github":"","link_paper":"","link_demo":"","details":"基于 RWKV-7 和开源数据集微调的心理咨询师模型，具备一定的心理问题问答能力","link_tutorial":""},{"id":20250610,"created_at":"2025-06-20T06:01:40.483177+00:00","name":"ARWKV","des":"RWKV 时间混合模块为核心的模型","category":"community","link_download":"https://huggingface.co/collections/RWKV-Red-Team/arwkv-7b-preview-01-679a55855cae3ab78d74f121","link_github":"https://github.com/yynil/RWKVInside","link_paper":"https://arxiv.org/abs/2501.15570","link_demo":"","details":"使用 RWKV-7 中的时间混合模块搭建的一种混合架构模型，通过高效蒸馏实现低成本训练。","link_tutorial":""},{"id":20250617,"created_at":"2025-06-20T05:42:54.325279+00:00","name":"rwkv-mobile-models ","des":"RWKV 的端侧量化模型","category":"community","link_download":"https://huggingface.co/mollysama/rwkv-mobile-models","link_github":"","link_paper":"","link_demo":"","details":"RWKV 的量化模型，主要用于端侧设备运行，有多种格式量化，目前主要支持 RWKV7-G1","link_tutorial":""},{"id":20250609,"created_at":"2025-06-20T03:59:07.366157+00:00","name":"RWKV-7-Arithmetic-0.1B","des":"基于 RWKV-LM 的加减法预训练模型","category":"community","link_download":"https://huggingface.co/shoumenchougou/RWKV-7-Arithmetic-0.1B","link_github":"","link_paper":"","link_demo":"","details":"使用合成数据基于 RWKV-LM 库预训练的 0.1B 加减法模型，下载链接包含模型和数据生成代码","link_tutorial":""},{"id":20250502,"created_at":"2025-05-12T06:37:29.65633+00:00","name":"rwkv7-g1-1.5b-Lonely-Neko","des":"基于 RWKV7-G1-1.5B 的中文情景对话模型","category":"community","link_download":"https://huggingface.co/Seikaijyu/rwkv7-g1-1.5b-Lonely-Neko","link_github":"","link_paper":"","link_demo":"","details":"rwkv7-g1-1.5b-Lonely-Neko 是一个单角色推理模型，拥有较为优秀的单角色rp能力。模型未训练任何nsfw语料，应用场景偏向日常闲聊。","link_tutorial":""},{"id":68,"created_at":"2025-04-01T08:07:32.56707+00:00","name":"WorldRWKV","des":"基于 RWKV 的全模态模型","category":"multimodal","link_download":"https://huggingface.co/WorldRWKV","link_github":"https://github.com/JL-er/WorldRWKV/","link_paper":"","link_demo":"","details":"WorldRWKV 目标是用纯 RWKV7 架构实现任意模态训练推理；现在可以使用 encoder 来任意切换模态的输入并输出文本。未来逐步实现端到端的跨模态推理，并且使用 RWKV7 来探索 World Model 的雏形。","link_tutorial":""},{"id":67,"created_at":"2025-04-01T07:01:23+00:00","name":"RWKV-TTS","des":"基于 RWKV 的音频模型","category":"multimodal","link_download":"https://huggingface.co/yueyulin/CosyVoice2-0.5B-RWKV-7-1.5B-Instruct-CHENJPKO","link_github":"https://github.com/yynil/RWKVTTS/","link_paper":"","link_demo":"","details":"RWKV-TTS 是基于 RWKV 架构的音频模型，传统的音频模型包含 VQ VAE 和 LLM 两部分，该项目专注于训练基于 RWKV 架构的 LLM 来替换之前的音频模型中的 LLM 部分。","link_tutorial":""},{"id":66,"created_at":"2024-11-01T05:46:56.192278+00:00","name":"Sudoku-RWKV","des":"数独专用 RWKV 模型","category":"community","link_download":"https://github.com/Jellyfish042/Sudoku-RWKV/blob/main/sudoku_rwkv_20241029.pth","link_github":"https://github.com/Jellyfish042/Sudoku-RWKV","link_paper":"","link_demo":"","details":"一个用于解决数独谜题的专用 RWKV 模型，仅 29M 参数，训练代码和训练数据制作脚本均已开源","link_tutorial":""},{"id":65,"created_at":"2024-10-09T09:51:45.430627+00:00","name":"RWKV-6 State","des":"RWKV State 文件","category":"official-fine-tuned","link_download":"https://hf-mirror.com/BlinkDL/temp-latest-training-models/tree/main","link_github":"","link_paper":"","link_demo":"","details":"作为 RNN 模型，RWKV 拥有固定大小的隐藏状态（State）。可通过加载 State 文件强化 RWKV 模型在特定任务的表现（类似于模型增强插件）。","link_tutorial":"https://rwkv.cn/news/read?id=343"},{"id":64,"created_at":"2024-10-09T09:50:35.163229+00:00","name":"RWKV-6-Jpn","des":"RWKV-6 日文微调模型","category":"official-fine-tuned","link_download":"https://hf-mirror.com/BlinkDL/rwkv-6-misc/tree/main","link_github":"","link_paper":"","link_demo":"","details":"RWKV-6-Jpn 系列日语模型基于 RWKV-6-World 模型微调而来，在日语任务和基准测试上表现良好。","link_tutorial":""},{"id":63,"created_at":"2024-10-09T09:49:40.797198+00:00","name":"RWKV-6-ChnNovel ","des":"RWKV-6 中文小说模型","category":"official-fine-tuned","link_download":"https://hf-mirror.com/BlinkDL/rwkv-6-misc/tree/main","link_github":"","link_paper":"","link_demo":"","details":"RWKV-6-ChnNovel 系列中文小说模型基于 RWKV-6-World 模型微调而来，在小说续写、小说扩写、角色扮演方面有非常好的效果。","link_tutorial":"https://rwkv.cn/news/read?id=4264"},{"id":61,"created_at":"2024-10-09T05:47:58.090448+00:00","name":"RWKV-6-3B-zh-cn-abnormal-text-review","des":"社区微调的文本审查模型","category":"community","link_download":"https://huggingface.co/Seikaijyu/RWKV-x060-World-3B-v2.1-zh-cn-abnormal-text-review-v0","link_github":"","link_paper":"","link_demo":"","details":"RWKV-6-3B-zh-cn-abnormal-text-review 是基于 RWKV6-v2.1-3B 基模微调的超小审查模型，主要用于审查文本中是否存在色情，涉政，不安全和辱骂内容。","link_tutorial":""},{"id":57,"created_at":"2024-10-09T05:47:58.090448+00:00","name":"Mod-RWKV","des":"RWKV 多模态内容审查模型","category":"community","link_download":"https://huggingface.co/modrwkv","link_github":"","link_paper":"","link_demo":"","details":"Mod-RWKV 是一个适用于各种内容模态（图像、视频、声音和文本）的内容审查模型，使用一个包含有害内容的数据集，对 RWKV 模型进行 SFT 微调和蒸馏。","link_tutorial":""},{"id":55,"created_at":"2024-10-09T05:47:58.090448+00:00","name":"RWKV-4-ABC-82M","des":"基于 RWKV-4 架构的 ABC 作曲模型","category":"music","link_download":"https://huggingface.co/BlinkDL/rwkv-4-music/tree/main","link_github":"","link_paper":"","link_demo":"","details":"RWKV-4-ABC-82M 是基于 RWKV-4 架构、使用大量 ABC 乐谱数据训练的作曲模型。可以在 chatRWKV 或 RWKV Runner 中运行 RWKV-4-ABC-82M 模型，生成 ABC 格式的乐谱。RWKV Runner 支持自动解析并播放模型生成的 ABC 乐谱。","link_tutorial":""},{"id":40,"created_at":"2024-10-09T05:47:58.090448+00:00","name":"RWKV-5-ABC-82M","des":"基于 RWKV-5 架构的 ABC 作曲模型","category":"music","link_download":"https://huggingface.co/BlinkDL/rwkv-5-music/tree/main","link_github":"","link_paper":"","link_demo":"","details":"RWKV-5-ABC-82M 是基于 RWKV-5 架构、使用大量 ABC 乐谱数据训练的作曲模型。可以在 chatRWKV 或 RWKV Runner 中运行 RWKV-5-ABC-82M 模型，生成 ABC 格式的乐谱。RWKV Runner 支持自动解析并播放模型生成的 ABC 乐谱。","link_tutorial":""},{"id":39,"created_at":"2024-10-09T05:47:58.090448+00:00","name":"RWKV-CLIP","des":"基于 RWKV 的视觉语言表示学习模型","category":"multimodal","link_download":"https://wisemodel.cn/models/deepglint/RWKV-CLIP/file","link_github":"https://github.com/deepglint/RWKV-CLIP","link_paper":"https://arxiv.org/abs/2406.06973","link_demo":"","details":"RWKV-CLIP 是一个 RWKV 驱动的视觉语言表示学习模型，该框架在多个下游任务中实现了最先进的性能，包括线性探测、零样本分类，以及零样本图像文本检索。","link_tutorial":""},{"id":38,"created_at":"2024-10-09T05:47:58.090448+00:00","name":"Diffusion-RWKV","des":"基于 RWKV 图像生成模型","category":"multimodal","link_download":"https://huggingface.co/feizhengcong/Diffusion-RWKV/tree/main","link_github":"https://github.com/feizc/Diffusion-RWKV","link_paper":"https://arxiv.org/abs/2404.04478","link_demo":"","details":"Diffusion-RWKV 是 RWKV 模型演变而来的架构，是一款应用于图像生成任务的 Diffusion 模型。Diffusion-RWKV 的独特优势体现在降低了空间聚合的复杂性，因此在处理高分辨率图像方面特别出色，从而消除了窗口化或分组缓存操作的必要性。","link_tutorial":""},{"id":54,"created_at":"2024-10-09T05:47:58.087759+00:00","name":"kgc_states_tuning","des":"适用于知识图谱构建任务的 RWKV State 文件","category":"community","link_download":"https://huggingface.co/yueyulin/kgc_states_tuning","link_github":"","link_paper":"","link_demo":"","details":"kgc_states_tuning 是一个社区微调的 RWKV State 文件，可用于根据给定的输入和格式提取三元组，进行知识图谱构建（knowledge graph construction）。该 State tuning 的训练数据覆盖多种 schema，具备一定的泛化能力，但是对于专业领域（如法律、医学等）的任务最好进行单独微调，欢迎提供数据让我们帮助微调。","link_tutorial":""},{"id":52,"created_at":"2024-10-09T05:47:58.087759+00:00","name":"Mobius-RWKV-r6-12B","des":"社区训练的 SFT RWKV-6 模型","category":"community","link_download":"https://huggingface.co/TimeMobius/Mobius-RWKV-r6-12B","link_github":"","link_paper":"","link_demo":"","details":"Mobius-RWKV-r6-12B 是基于 RWKV-6 架构、在一定量数据上预训练的 pretrain+sft 模型。与预训练 RWKV 模型相比，Mobius-RWKV-r6-12B 模型的中文表现会更好、稳定支持16K上下文长度，且支持 function call。","link_tutorial":""},{"id":51,"created_at":"2024-10-09T05:47:58.087759+00:00","name":"rwkv6_crossencoder","des":"基于 RWKV-6 的 Cross-Encoder","category":"community","link_download":"https://huggingface.co/yueyulin/rwkv6_crossencoder","link_github":"","link_paper":"","link_demo":"","details":"rwkv6_crossencoder 这是基于 RWKV-6 架构的 Cross-Encoder （交叉编码器） ，既可以用于生成嵌入向量的 embedding 任务，也可以用于重排序查询/上下文检索的 Rerank 任务。","link_tutorial":""},{"id":50,"created_at":"2024-10-09T05:47:58.087759+00:00","name":"rwkv6_emb_4k_base","des":"RWKV-6 embedding 模型","category":"community","link_download":"https://huggingface.co/yueyulin/rwkv6_emb_4k_base","link_github":"","link_paper":"","link_demo":"","details":"rwkv6_emb_4k_base 是基于 RWKV-6 的 embedding 嵌入模型（使用中文查询/上下文数据进行微调的 Bi-Encoder），主要用于将文本转换成嵌入向量。","link_tutorial":""},{"id":48,"created_at":"2024-10-09T05:47:58.087759+00:00","name":"RWKV-4-MIDI-560M","des":"基于 RWKV-4 架构的 MIDI 作曲模型","category":"music","link_download":"https://huggingface.co/BlinkDL/rwkv-4-music/tree/main","link_github":"","link_paper":"","link_demo":"","details":"RWKV-4-MIDI-560M 是基于 RWKV-4 架构、使用大量 ABC 乐谱数据训练的作曲模型。可以在 chatRWKV 或 RWKV Runner 中运行 RWKV-4-MIDI-560M 模型，生成 MIDI 格式的乐谱。RWKV Runner 支持自动解析并播放模型生成的 MIDI 乐谱。","link_tutorial":""},{"id":46,"created_at":"2024-10-09T05:47:58.087759+00:00","name":"RWKV-5-MIDI-560M","des":"基于 RWKV-5 的 MIDI 作曲模型","category":"music","link_download":"https://huggingface.co/BlinkDL/rwkv-5-music/tree/main","link_github":"","link_paper":"","link_demo":"","details":"RWKV-5-MIDI-560M 是基于 RWKV-5 架构、使用大量 MIDI 乐谱数据训练的作曲模型。可以在 chatRWKV 或 RWKV Runner 中运行 RWKV-5-MIDI-560M 模型，生成 MIDI 格式的乐谱。RWKV Runner 支持自动解析并播放模型生成的 MIDI 乐谱。","link_tutorial":""},{"id":42,"created_at":"2024-10-09T05:47:58.087759+00:00","name":"Vision-RWKV","des":"基于 RWKV 的视觉任务增强模型","category":"multimodal","link_download":"https://huggingface.co/OpenGVLab/Vision-RWKV/tree/main","link_github":"https://github.com/OpenGVLab/Vision-RWKV","link_paper":"https://arxiv.org/abs/2403.02308","link_demo":"","details":"Vision-RWKV（VRWKV） 是在 RWKV 语言模型的基础上改进而成的适用于视觉任务的模型。VRWKV 可以作为 ViT 的低成本替代方案，它在图像分类任务中优于 ViT，在处理高分辨率输入时速度更快，内存效率更高。","link_tutorial":""},{"id":35,"created_at":"2024-10-09T05:47:58.087759+00:00","name":"VisualRWKV-6","des":"基于 RWKV-6 架构的图像处理模型","category":"multimodal","link_download":"https://huggingface.co/howard-hou/visualrwkv-6","link_github":"https://github.com/howard-hou/VisualRWKV","link_paper":"https://arxiv.org/abs/2406.13362","link_demo":"https://huggingface.co/spaces/howard-hou/VisualRWKV-Gradio-1","details":"VisualRWKV 是 RWKV 语言模型的图像处理增强版本，使 RWKV 模型能够处理各种图像任务。大量实验表明，与基于 Transformer 的模型（如 LLaVA-1.5）相比，VisualRWKV 在各种基准测试中实现了具有竞争力的性能。","link_tutorial":""}]}