Transformers Model Integration Contracts
Hugging Face 给 Transformers 定了套模型接入契约，想给库提交新模型架构的人可以看看规则怎么变了。
Hugging Face 在官方博客上介绍了 Transformers Model Integration Contracts（MIC）。MIC 是一套模型集成契约，用于约束新模型接入 Transformers 库的方式。该机制面向向 Transformers 提交新模型架构的开发者与维护者。
三个后端、五千多个仓库被扒了一遍配置，想知道大家平时真的调哪些推理参数、哪些基本没人碰，看这篇就够了。
A single-pass router that answers 64 routing questions per request can cut inference overhead for LLM gateway deployments compared with multi-call routing pipelines. Builders running self-hosted model fleets can plug the Apache-2.0 checkpoints straight into a Transformers workflow and tune the size-performance trade-off from 0.6B to 27B. It also gives teams an open-source alternative to proprietary model-routing services.
Speaker diarization has been a missing piece in voice agent stacks, where most demos hide the inability to tell callers apart. Because the model works on one-second chunks, developers can plug it into streaming pipelines without waiting for turn ends. The commercial-friendly license and day-zero Transformers support make it a practical alternative to closed speech APIs for anyone building voice robots or assistants.
Developers who serve GGUF-quantized open-source models can now run them directly through Transformers with llama.cpp-level speed, removing a common reason to keep two separate inference stacks.
Hugging Face 把 Transformers 模型直接跑进 vLLM，性能不输手写，以后搞推理不用重复造轮子了。
