TrustRadius: an HG Insights company

What is InclusionAI Ling?

The Ling Model Family is an open-weight Mixture-of-Experts (MoE) large language model series developed and open-sourced by inclusionAI (Ant Group). Distributed via Hugging Face and ModelScope, the model family provides foundational language generation, reasoning, and coding capabilities across base and instruction-tuned (chat) checkpoints.

Parameter Sizes & Configurations

  • Ling-lite: Features 16.8 billion total parameters with 2.75 billion activated parameters per token and supports context lengths up to 128,000 tokens.
  • Ling-plus: Designed for higher-capacity workloads with 290 billion total parameters and 28.8 billion activated parameters per token, supporting a 64,000-token context window.
  • Ling-coder-lite: A code-specialized variant built on the 16.8B total / 2.75B activated parameter architecture, configured with a 16,000-token context window.

Deployment & Infrastructure

Because weights are publicly released, Ling models are deployed on self-managed or customer-controlled inference infrastructure rather than through a vendor-managed API. The models support deployment through:

  • Hugging Face Transformers: For direct Python integration and model loading.
  • vLLM: For offline batched processing or online OpenAI-compatible API endpoints, using YaRN RoPE scaling to handle extended context windows.
  • MindIE Framework: For execution on Huawei Ascend NPU hardware architectures.

Ling serves as the foundational text-generation layer within inclusionAI’s open-source ecosystem, which also encompasses the Ring series for specialized reasoning and the Ming family for multimodal perception and generation.

Categories & Use Cases

Technical Details

Technical Details
Mobile ApplicationNo

FAQs

What is InclusionAI Ling?
The Ling Model Family is an open-weight Mixture-of-Experts (MoE) large language model series developed and open-sourced by inclusionAI (Ant Group). Distributed via Hugging Face and ModelScope, the model family provides foundational language generation, reasoning, and coding capabilities across base and instruction-tuned (chat) checkpoints.