Liquid AI released speculative decoding DSpark draft models for LFM2.5-1.2B-Instruct, LFM2.5-2.6B, and LFM2.5-8B-A1B, delivering up to 3.18x speedups on GPUs and 2.87x on edge devices.

Key Details

• Released DSpark draft model checkpoints for LFM2.5-1.2B-Instruct, LFM2.5-2.6B, and LFM2.5-8B-A1B in Safetensors and GGUF formats
• Achieves up to 3.18x throughput improvement on H100 GPUs and up to 2.87x on M4 Max MacBooks
• Reduces function-calling latency by an average of 57% for LFM2.5-2.6B
• Features day-one upstream integration with llama.cpp and SGLang inference engines
• Draft models contain approximately 300M parameters across 5 decoder layers with a block size of 9

In This Story