我们已开源 Ling-3.0 系列,这是我们迄今为止最高效的语言基础模型家族。为支持研究和社区驱动的创新,我们发布了训练过程中的一系列检查点,具体如下:
| 模型 | 预训练 | 中期训练 | 合并(即 WSM) |
|---|---|---|---|
| Ling-3.0-tiny | Ling-3.0-tiny-base-30T | Ling-3.0-tiny-base-midtrain | Ling-3.0-tiny-base |
| Ling-3.0-flash | Ling-3.0-flash-base-30T | Ling-3.0-flash-base-midtrain | Ling-3.0-flash-base |
这些检查点对应训练过程的不同阶段:
预训练检查点已完成大规模预训练,但尚未经过中期训练、WSM 合并(或学习率衰减)或后训练。中期训练检查点已完成中期训练,但尚未经过 WSM 合并(或学习率衰减)或后训练。合并检查点已在中期训练检查点的基础上完成 WSM 合并(或学习率衰减),但尚未经过后训练。
发布这些检查点是为了支持继续预训练、微调和进一步研究。关于后训练模型,请参阅 Ling-3.0-tiny 和 Ling-3.0-flash。
稀疏 MoE 架构:128 个路由专家,每个 token 仅激活 8 个路由专家和 1 个共享专家。这使得模型在每次 token 仅激活 1.3B 参数的情况下仍具备广泛的能力;原生混合线性注意力:Ling-3.0 系列从预训练一开始就采用原生混合线性注意力架构,将 KDA 与 Gated MLA 相结合,以实现长上下文输入的高效处理。Warmup-Stable 与合并:我们用加权检查点合并取代了传统的学习率衰减。通过消除衰减阶段,我们的基础模型更适合持续预训练和动态数据扩展,同时支持离线探索不同的衰减配置,而无需为每种策略重新运行昂贵的实验。无缝扩展:Ling-3.0-tiny-base 和 Ling-3.0-flash-base 采用相同的训练方案,使社区能够先在 Ling-3.0-tiny-base 上进行实验,然后将验证过的训练策略扩展到更大的 Ling-3.0-flash-base。
| 模型类型 | 基础(预训练的最终检查点) |
|---|---|
| 架构 | 混合线性 MoE |
| 参数量规模 | 总参数量 7.9B,激活参数量 1.3B |
| Transformer 层数 | 18 层 KDA + 6 层 Gated MLA(3:1) |
| 稠密层数量 | 1 |
| 路由专家数量 | 128 |
| 共享专家数量 | 1 |
| 激活专家数量 | 8 |
| 注意力头数 | 16 |
| 隐藏层大小 | 1536 |
| 专家中间层大小 | 512 |
| 稠密层中间大小 | 4608 |
| 词表大小 | 157,184 |
为系统评估基础模型的能力,我们使用了一套全面的评测基准,覆盖知识、编程、数学、推理、多语言理解以及长上下文理解等多个关键领域。预训练基础检查点(即 Ling-3.0-tiny-base)的性能对比如下:
推荐使用场景:
继续预训练、中期训练、面向领域适配的监督微调、偏好优化与 RL 后训练、知识蒸馏研究、长上下文与 MoE 系统研究
不建议直接用于:
直接面向终端用户的聊天部署、未经额外对齐与评估的安全关键型应用、未经后训练与任务特定验证的生产环境使用
使用方法
微调示例请参考我们的 ling-cookbook。
常见问题
如有任何疑问,欢迎发起讨论。
许可证
本模型基于 MIT 许可证发布。
该模型尚未由任何推理服务提供商部署。🙋如需提供商支持,请提出请求%20to%20be%20supported%20by%20Inference%20Providers.%0A%0A(optional)%20Which%20providers%20are%20you%20interested%20in%3F%20(Novita%2C%20Hyperbolic%2C%20Together%E2%80%A6)%0A)
包含 inclusionAI/Ling-3.0-tiny-base-30T 的模型集合
Ling 3.0 •13 个项目•17 分钟前更新• 15
inclusionAI/Ling-3.0-tiny-base-30T 的论文
论文 • 2507.17634 •发布于 2025 年 7 月 23 日• 2
We have open-sourced the Ling-3.0 series, our most efficient language foundation model family to date. To support research and community-driven innovation, we are releasing a collection of checkpoints during the training process as following:
| Model | Pre-trained | Mid-trained | Merged (i.e., WSM) |
|---|---|---|---|
| Ling-3.0-tiny | Ling-3.0-tiny-base-30T | Ling-3.0-tiny-base-midtrain | Ling-3.0-tiny-base |
| Ling-3.0-flash | Ling-3.0-flash-base-30T | Ling-3.0-flash-base-midtrain | Ling-3.0-flash-base |
These checkpoints correspond to different stages of the training process:
Pretrained checkpoint have completed large-scale pretraining but have not undergone mid-training, WSM merging (or learning-rate decay), or post-training. Mid-trained checkpoint have completed mid-training but have not undergone WSM merging (or learning-rate decay) or post-training. Merged checkpoints have undergone WSM merging (or learning-rate decay) based on the mid-training checkpoints but have not undergone post-training.
These checkpoints are released to support continued pretraining, fine-tuning, and further research. For the post-trained model, please see Ling-3.0-tiny and Ling-3.0-flash.
Sparse MoE architecture: 128 routed experts, with only 8 routed experts and 1 shared expert activated per token. This enables broad model capabilities while activating just 1.3B parameters per token; Native hybrid linear attention: Ling-3.0 series adopt a native hybrid linear attention architecture from the very start of pretraining by combining KDA with Gated MLA to enable efficient processing of long-context inputs. Warmup-Stable and Merge: We replace conventional learning-rate decay with weighted checkpoint merging. By eliminating the decay phase, our Base Model is better suited for continual pretraining and dynamic data expansion, while enabling offline exploration of different decay profiles without rerunning costly experiments for each strategy. Scale Seamlessly: Ling-3.0-tiny-base and Ling-3.0-flash-base share the same training recipe, enabling community to experiment on the Ling-3.0-tiny-base first and then scale validated training strategies to the larger Ling-3.0-flash-base.
| Model Type | Base (final checkpoint of pre-training) |
|---|---|
| Architecture | Hybrid-linear MoE |
| Parameter Scale | Totoal 7.9B, Activated 1.3B |
| Transformer Layers | 18 KDA + 6 Gated MLA (3:1) |
| Number of Dense Layers | 1 |
| Number of Routed Experts | 128 |
| Number of Shared Experts | 1 |
| Number of Activated Experts | 8 |
| Attention Heads | 16 |
| Hidden Size | 1536 |
| Expert Intermediate Size | 512 |
| Dense Intermediate Size | 4608 |
| Vocabulary Size | 157,184 |
To systematically assess the capabilities of the base model, we use a comprehensive benchmark suite covering several key domains, including knowledge, coding, mathematics, reasoning, multilingual understanding, and long-context comprehension. The performance of the pretrained base checkpoint, i.e., Ling-3.0-tiny-base, is compared below:
Recommended use cases:
Continued pre-training Mid-training Supervised fine-tuning for domain adaptation Preference optimization and RL post-training Distillation research Long-context and MoE systems research
Not recommended as-is for:
Direct end-user chat deployment Safety-critical applications without additional alignment and evaluation Production use without post-training and task-specific validation
Usage
For fine-tuning examples, please refer to our ling-cookbook.
FAQ
If you have any question, please feel free to add a discussion.
License
This model is released under the MIT License.
This model isn't deployed by any Inference Provider.🙋Ask for provider support%20to%20be%20supported%20by%20Inference%20Providers.%0A%0A(optional)%20Which%20providers%20are%20you%20interested%20in%3F%20(Novita%2C%20Hyperbolic%2C%20Together%E2%80%A6)%0A)
Collection including inclusionAI/Ling-3.0-tiny-base-30T
Ling 3.0 •13 items•Updated 17 minutes ago• 15
Paper for inclusionAI/Ling-3.0-tiny-base-30T
Paper • 2507.17634 •Published Jul 23, 2025• 2