Liquid AI releases DSpark draft models, boosting inference speed by up to 3.18x
JosephJacks_ · x · 2026-08-21
Liquid AI released DSpark draft models for the LFM2.5 family, utilizing speculative decoding to trade minimal memory increases for significant speedups without compromising output quality.
- Mechanism: A lightweight draft model proposes candidate token blocks, verified by the target model in a single forward pass.
- Benchmarks: LFM2.5-8B-A1B achieved a 3.18x throughput increase on H100 (428 → 1362 tok/s) on MATH500. On M4 Max, the 1.2B model saw a 2.87x boost on HumanEval.
Related event: Liquid AI Releases DSpark Draft Model, Speeding Decoding up to 3.18x(5 posts)→
More from Models
- Hash Rate Ep. 189: Jev models on Bittensor subnet SN92, decentralized training vs traditional LLMs — markjeffrey · 2026-10-17
- 'Remove AI tics and mannered prose': a one-line prompt fix for Argon's writing — arjunrajlab · 2026-10-17
- Blind test of 200 real image edits: nano banana pro loses to its cheaper base model — ravisparikh · 2026-10-17
- Dev ditches Gemini 4 Argon for Grok 4.5, which eats 96.6% of his Cursor tokens — gaganghotra_ · 2026-10-17
- LMArena opens Code Arena WebDev for blind head-to-head model coding battles — arena · 2026-10-17
- Unverified: GPT-6 Pro said to launch parallel research subagents, 50-100 weekly messages — TheMoonMidas · 2026-10-17