Up to 3.2x Faster Inference with LFM2.5-DSpark

Liquid AI released DSpark draft checkpoints for LFM2.5-1.2B-Instruct, LFM2.5-2.6B, and LFM2.5-8B-A1B, adding speculative decoding with ~300M-parameter drafters. Throughput reaches 3.18x on an H100 and 2.87x on an M4 Max MacBook Pro. Greedy output stays identical to baseline, and multi-tool function-calling latency drops 57% for LFM2.5-2.6B.