Unsloth's new release supports training text models on CPU alone and speeds up generation

Unsloth has released v0.1.914 beta. Users without a GPU can now fine-tune text models on the CPU straight from the Train page. Generation is faster too: some single-GPU Llama and Qwen3 models run faster on NVIDIA GPUs under Linux or WSL, and in Unsloth's own A100 test a 4-bit Qwen3-8B went from 25.7 to 99.1 tokens per second, about 3.9 times faster. The release also adds multi-GPU training settings: models too large for one GPU can be split by layer across several GPUs, and a setting called UNSLOTH_PARALLEL combines several ways of spreading training over multiple GPUs. The interface gains Glass themes, custom greetings and backgrounds, and a chat navigator for previewing, bookmarking and jumping back to earlier messages.

Written by dalao.ai: the title, summary and why it matters are our own words. The original reports are the record; their publishers keep authorship.