181 Tok/sec running on my shitty laptop GPU. On a good GPU this can push 400tok/s
It's not the best AI but it can code and run Hermes lightning fast.
I included my recipe I've been working on all day. Enjoy buddies

GitHub
GitHub - barrydeen/lfm3-dspark: Run OpenIntelligenceNet/LFM3-1.7B-Uncensored with lossless DSpark speculative decoding on vLLM โ OpenAI-compatible API, ~1.9x decode speedup
Run OpenIntelligenceNet/LFM3-1.7B-Uncensored with lossless DSpark speculative decoding on vLLM โ OpenAI-compatible API, ~1.9x decode speedup - ba...