📦 MLX-LM – v0.26.3
🎉 AI Enthusiasts, Heads Up! MLX LM Just Got a Big Update (v0.26.3)! 🎉
Get ready to unleash even more power with MLX LM! The latest version, v0.26.3, is here and packed with improvements. Here’s a quick look at what’s new:
- Bug Squashing: Several bugs have been fixed, including a streaming token error, a NameError, and response type errors, making your experience smoother.
- New CLI Options: You can now use `–trust-remote-code` and `–confirm-run-unsafe-code` for more control. There’s also a validation set for DWQ.
- Model Mania! 🤩 Huge news: Support for the GPT-OSS and Hunyuan V1 Dense models has been added! Plus, exciting additions for GPT-OSS like LoRA, Alternating Attention, and MoE support.
- Under the Hood Improvements: A Jensen-Shannon divergence loss kernel has been added, and GPT-OSS is now routed to fused SDPA for improved performance.
- Big Thanks! A warm welcome to our new contributors: @zenyr, @snellingio, @emmanuel-ferdman, @dojoteef, @vsabolcec, and @Shashikant86 for their awesome contributions!
Check out the full changelog for all the details: https://github.com/oobabooga/text-generation-webui/blob/main/CHANGELOG.md#v0262v0263
Get the latest release here: https://github.com/ml-explore/mlx-lm/releases/tag/v0.26.3
🔗 https://github.com/ml-explore/mlx-lm/releases/tag/v0.26.3