LFM3 1.7B impresses in local test: 1.7GB Q8 quant hits 329 tok/s on RTX 3090 Ti
Repulsive_Initial308 · reddit · 2026-11-04
A Reddit user tested LiquidAI's tiny LFM3 1.7B model and found it surprisingly capable.
- Running the Q80 GGUF quant (1.71 GiB) via llama.cpp on an RTX 3090 Ti: 20,648 tokens/s at pp512 prefill and 329 tokens/s at tg128 generation.
- The 2B VL variant also performed well, but has half the context length of the 1.7B model.
- The fast results prompted the author to look into LiquidAI's omni model for video understanding, asking for an easy way to feed video files or YouTube links into chat.
Useful benchmark data point for anyone experimenting with small local models.
More from Models
- Editing 0.13% of weights, no training, fixes LightOnOCR-2-1B blank-page hallucination — vanstriendaniel · 2026-11-04
- antirez: benchmark games trivially guessable from output biases; Astra misses Cannon — antirez · 2026-11-04
- antirez replicates MIT weight-rewriting paper, finds visible radial patterns in first-layer weights — antirez · 2026-11-04
- 0.6B post-trained model fails MATH: tiny weight defect makes it sampling-sensitive — lateinteraction · 2026-11-04
- $500 Claude plan burns a week of credits every 36 hours, user complains — StephanSturges · 2026-11-04
- FireFlash coding agent hits 46.2% on Terminal-Bench 4.0 at 30x lower cost — VioletNPeng · 2026-11-04