Liquid AI's LFM2.5-2.6B brings on-device agents to edge hardware
A 2.6B-parameter model trained for tool use and multi-step reasoning, matching performance of models 4x larger while running at 220 tokens/sec on consumer hardware.
#model-optimization
LoRA's Dominance in Fine-Tuning Masks a Crowded Field of Alternatives
Hugging Face data shows LoRA controls 98% of PEFT implementations, but emerging techniques like DoRA and LoHa challenge its monopoly on parameter-efficient adaptation.
Hugging Face Cuts RL Training Sync Overhead by 98% With Sparse Delta Weights
A new TRL protocol reduces per-step model synchronization from terabytes to tens of megabytes by shipping only changed parameters across distributed training pipelines.