Liquid's LFM3-1.7B punches above its weight in local long-context tests
mumblerit · reddit · 2026-11-01
A Reddit user tested Liquid's LFM3-1.7B locally: Q80 GGUF via llama.cpp on ROCm hits 4.1k t/s prefill and 78 t/s decode on a Strix Halo iGPU, or 7.1k/151 t/s tensor-split across two 7900 cards. Notably coherent at 32k-64k prompts — far better than a 1.7B model deserves, making it a solid pick for compression/summarization.
More from Infra
- Wan 2.2 color pulse fix and other hard-won tips from rendering a music video on one AMD GPU — wilppuse · 2026-11-02
- Chamath: Power is THE binding constraint in AI — and hyperscalers are good at this too — BenBajarin · 2026-11-02
- Drex 2.0 runs fully local on a Mac: 21GB weights, 113ms per decision — testingcatalog · 2026-11-02
- "AWS faced the same skepticism in 2008" — analyst pushes back on AI datacenter bubble fears — BenBajarin · 2026-11-02
- $100M Neolab tries to sublease its GPUs; compute glut's debt risk called 'tip of the iceberg' — himanshustwts · 2026-11-02
- 'Rampocalypse Now' meme mocks the DRAM price surge as an apocalypse — crantob · 2026-11-01