Liquid AI Releases LFM2.5-VL-3B-DSpark: Speculative Decoding for Vision-Language Models With Up to 3.13x Faster Decoding
Liquid AI released a 280M drafter that speeds LFM2.5-VL-3B decoding by up to 3.13x.
Liquid AI released LFM2.5-VL-3B-DSpark, an experimental 279.5-million-parameter speculative-decoding drafter for its LFM2.5-VL-3B vision-language model. The four-layer drafter proposes tokens that the target verifies, adding about 8.9% parameters without changing greedy outputs. Liquid AI reports decode speedups up to 3.13x on an Apple M5 Max and 2.66x on an NVIDIA H100, with smaller end-to-end gains because image encoding and prefill are unchanged. Weights are on Hugging Face in Safetensors and GGUF, with support in SGLang, MLX-VLM, and llama.cpp, under a license that allows free commercial use only for companies under $10 million in annual revenue.
- The 279.5M drafter adds about 8.9% parameters and leaves greedy outputs unchanged.
- Decode speed reached up to 3.13x on an M5 Max and 2.66x on an H100.
- Weights are on Hugging Face in Safetensors and GGUF under the LFM Open License.
- Image encoding and prefill limit end-to-end gains, especially on edge devices.
- Free commercial use is limited to companies under $10 million annual revenue.
- Vendors
- Liquid AI
- Organizations
- Liquid AIHugging Face
- AI models
- LFM2.5-VL-3BLFM2.5-VL-3B-DSpark
From the original article
4 min read at marktechpost.com
Liquid AI has released LFM2. 5-VL-3B-DSpark, a 279. 5M-parameter draft model that brings speculative decoding to its LFM2.
By Asif Razzaq · marktechpost.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.