Hugging Face Transformers v5.11.0 adds DiffusionGemma and DeepSeek-V3.2

Hugging Face's Transformers library has released version 5.11.0, introducing new models like DiffusionGemma and DeepSeek-V3.2. DiffusionGemma utilizes an encoder-decoder architecture with multi-canvas sampling for faster inference, while DeepSeek-V3.2 incorporates a novel sparse attention mechanism for improved long-context efficiency. The update also includes enhancements to kernel integration, parallelization, and various bug fixes across the library. AI

IMPACT Integrates new models and efficiency techniques, potentially improving performance for users of the Transformers library.

RANK_REASON This is a software library release with new model integrations and technical improvements, not a frontier model release from a major lab. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Transformers — Releases →

• DeepSeek Sparse Attention
• DeepSeek-V3.2
• DiffusionGemma
• FalconMamba
• Hugging Face
• NemotronH
• Qwen2.5-VL
• Qwen2-VL
• Qwen3-VL
• Transformers
• v5.11.0
• Zamba2

• model release
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →