OpenAI's GPT-OSS models introduce several efficiency upgrades for transformers, including MXFP4 quantization and specialized kernels that enhance performance during model loading and execution. The updates allow for faster inference and fine-tuning while maintaining compatibility across major models in the transformers library. Additionally, community-contributed kernels are integrated to streamline usage and performance optimization.
SGLang has integrated Hugging Face transformers as a backend, enhancing inference performance for models while maintaining the flexibility of the transformers library. This integration allows for high-throughput, low-latency tasks and supports models not natively compatible with SGLang, streamlining deployment and usage. Key features include automatic fallback to transformers and optimized performance through mechanisms like RadixAttention.