Announcement
Axolotl v0.21.0 adds 501B MoE fine-tuning on eight H200 GPUs
The release post also lists native diffusion LM training, multimodal DPO, KTO and GRPO, and CoreWeave cloud training.
TLDR
A post announcing Axolotl v0.21.0 says it includes 501B MoE fine-tuning on eight H200 GPUs, native diffusion LM training, multimodal DPO, KTO and GRPO, and CoreWeave cloud training. It claims eager LoRA is 8% faster and notes that v0.20.0 added GGUF exports, NVFP4 LoRA and long-context training.
Combined views
344
1 Source, first seen ago
