⚡ Performance and Efficiency Benchmarks

This section reports the performance on NPU with FastFlowLM (FLM).

Note:

  • Results are based on FastFlowLM v0.9.19.
  • Under FLM’s default NPU power mode (Performance)
  • Test system spec: AMD Ryzen™ AI 7 350 (Krakan Point) with 32 GB DRAM.
  • Newer versions may deliver improved performance.

🚀 Decoding Speed (TPS, or Tokens per Second, starting @ different context lengths)

Model Hardware 1k 2k 4k 8k 16k 32k Model
LFM2-1.2B NPU (FLM) 41.5 40.6 39.6 37.1 33.3 28 LFM2-1.2B

🚀 Prefill Speed (TPS, or Tokens per Second, with different prompt lengths)

Model Hardware 1k 2k 4k 8k 16k 32k Model
LFM2-1.2B NPU (FLM) 1384 1781 2038 1905 1503 1024 LFM2-1.2B