safetensors

safetensors 0.8.0 release brings major performance improvements: direct copy into Metal MTLBuffers with dlpack for 2-3x faster loading and OOM fix on macOS, plus GIL-free serialization for faster multi-file saves.

A fine-tuned uncensored version of the Qwen model (Qwen3.6-35B-A3B) with MTP support and APEX quantization, tested stable at 200k context and recommended for use in LM Studio.

Hugging Face flagged a safetensors file as unsafe, confusing users who question the policy.

The MiniMax M2.7 model has been fine-tuned into an uncensored variant, 'ultra uncensored heretic', with very low refusal rates (4/100). Available in Safetensors and GGUF formats on HuggingFace.

Community release of Qwen3.6 35B A3B uncensored variant with full 19 MTP tensors preserved, available in multiple formats including Safetensors, GGUF, NVFP4 and GPTQ-Int4.

Safetensors has officially joined the PyTorch Foundation under the Linux Foundation to establish a vendor-neutral governance structure, while maintaining its status as the default model format for the Hugging Face Hub.