Explore this comparison’s ranking

Comparison Counted

Which engine runs LFM2.5 350M best on iPhone 12 · 4 GB?

Device
iPhone 12 · 4 GB (iPhones)
OS
iOS 27
Model
LFM2.5 350M, 350M parameters
Weights
4-bit
Workload
Archive 256 input / 128 output: 256 tokens in, 128 out
Decoding
Greedy, batch 1, context 4,096
Power
Plugged in

Speed: Decode speed: tokens generated per second.

Which engine runs LFM2.5 350M best on iPhone 12 · 4 GB? Ranked by Speed.
#EngineDecode and First token
1uzu, an approximate comparison, see details Mirai Decode: 130.8 tok/s 384 ms first token
2llama.cpp, an approximate comparison, see details ggml-org Decode: 56.0 tok/s 314 ms first token
Not ranked in this view
Not rankedExecuTorch PyTorch (Meta) Not counted Its output check failed: 5/8 of the checks the model passes at BF16; needs 7 and the mandatory ones (needle-start, needle-end). Left out, failed by every BF16 configuration: arithmetic, sort, word, continuation-128.
details

Speed Decode speed: tokens generated per second.

Score: 100 is the best result in every comparison an engine entered; 80 is 80% of the best, on average. Only complete experiments with a passed output check count. How rankings work

† uzu: approximate comparison. The experiments use different effective KV cache precision: fp32, f16, uzu default. The experiments here use different 4-bit schemes: 8da4w, Q4_K_M, mirai-m.

† llama.cpp: approximate comparison. The experiments use different effective KV cache precision: fp32, f16, uzu default. The experiments here use different 4-bit schemes: 8da4w, Q4_K_M, mirai-m.

Every experiment

Medians over valid trials, with the lowest and highest beneath. Eligible experiments are complete and passed their output check. The ranking above selects each engine's configuration for this objective when its required metrics are measured.

Every experiment in this comparison
ExperimentRanking eligibilityDecode tok/s, higher is betterFirst token ms, lower is betterMemory GiB, lower is betterPrefill tok/s, higher is betterValid trials
uzu 0.6.17 · Metal · mirai-m Eligible, approximately: The experiments use different effective KV cache precision: fp32, f16, uzu default. The experiments here use different 4-bit schemes: 8da4w, Q4_K_M, mirai-m.130.8 range126.8–131.2384 range379–400Not measured786.2 range777.4–795.56of 6 planned
llama.cpp b12110 · Metal · Q4_K_M Eligible, approximately: The experiments use different effective KV cache precision: fp32, f16, uzu default. The experiments here use different 4-bit schemes: 8da4w, Q4_K_M, mirai-m.56.0 range52.1–57.4314 range312–317Not measured818.5 range809.3–823.76of 6 planned
ExecuTorch 1.6.0 · CPU · 8da4w Complete Its output check failed: 5/8 of the checks the model passes at BF16; needs 7 and the mandatory ones (needle-start, needle-end). Left out, failed by every BF16 configuration: arithmetic, sort, word, continuation-128.121.9 range110.3–122.8481 range469–582Not measured532.2 range439.9–547.06of 6 planned

Every trial, metric by metric

The line runs from an experiment's lowest to its highest valid trial, the dot is the median and each tick is one trial. Every scale starts at zero.

Caveats

  • Publisher cards identify the base model but do not always pin the exact export source checkpoint revision.
  • Engines may differ in quantization scheme, effective KV cache precision, kernels, and activations.

About this comparison

Worked out from its experiments: every experiment here shares the device, the OS major version, the power state, the base model, the quantization class and the workload. Measured under methodology v0.6. Comparison id glenn-iphone-12--ios-27--lfm2.5-350m--4-bit--archive-256-128.