News
Liquid AI Releases LFM2.5-DSpark Inferential Decoding Model
Liquid AI released the LFM2.5-DSpark speculative decoding model, providing matching checkpoints for 1.2B-Instruct, 2.6B, and 8B-A1B; vendor tests claim it can accelerate inference by up to 3.2x.
Liquid AI released the LFM2.5-DSpark speculative decoding model, providing checkpoints for three LFM2.5 models.
The initial rollout covers 1.2B-Instruct, 2.6B, and 8B-A1B. Liquid AI tests claim inference speed improvements of up to 3.2x, but this result comes from vendor testing, and actual gains will vary depending on hardware, batch, cue length, and inference framework.
source:Liquid AI official article。