Explore this comparison’s ranking

Comparison Counted

Which engine runs LFM2.5 350M best on M5 Max · 64 GiB?

Device
M5 Max · 64 GiB (Apple silicon Macs)
OS
macOS 27
Model
LFM2.5 350M, 350M parameters
Weights
8-bit
Workload
Archive 2048 input / 128 output: 2,048 tokens in, 128 out
Decoding
Greedy, batch 1, context 4,096
Power
Plugged in

Memory: Peak memory while generating: the less, the better.

Which engine runs LFM2.5 350M best on M5 Max · 64 GiB? Ranked by Memory.
#EngineMemory and Decode
1Ollama, an approximate comparison, see details Ollama Memory: 0.61 GiB 599.8 tok/s decode
2llama.cpp, an approximate comparison, see details ggml-org Memory: 0.61 GiB 592.1 tok/s decode
3uzu, an approximate comparison, see details Mirai Memory: 0.70 GiB 776.1 tok/s decode
4Magnitude, an approximate comparison, see details Magnitude Memory: 1.04 GiB 747.9 tok/s decode
5LiteRT-LM, an approximate comparison, see details Google AI Edge Memory: 1.25 GiB 198.9 tok/s decode
6Rapid-MLX, an approximate comparison, see details Rapid-MLX contributors Memory: 2.05 GiB 712.2 tok/s decode
7MLX LM, an approximate comparison, see details Apple (ml-explore) Memory: 2.09 GiB 702.2 tok/s decode
details

Memory Peak memory while generating: the less, the better.

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

† Ollama: approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default.

† llama.cpp: approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default.

† uzu: approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 8-bit in another scheme: mirai-l, where most use Q8_0.

† Magnitude: approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its phases were timed by the harness clock around the engine API, so they include transport that engine counters leave out.

† LiteRT-LM: approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 8-bit in another scheme: litert-int8, where most use Q8_0.

† Rapid-MLX: approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 8-bit in another scheme: mlx-affine-8bit-g64, where most use Q8_0.

† MLX LM: approximate comparison. The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 8-bit in another scheme: mlx-affine-8bit-g64, where most use Q8_0.

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
Ollama 0.40.13 · Metal · Q8_0 Eligible, approximately: The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default.599.8 range586.3–603.961.5 range61.5–61.70.61 range0.61–0.6134,935 range34,835–35,1086of 6 planned
llama.cpp b12110 · Metal · Q8_0 Eligible, approximately: The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default.592.1 range568.8–612.163.4 range61.9–70.30.61 range0.61–0.6132,938 range29,505–33,7636of 6 planned
uzu 0.6.17 · Metal · mirai-l Eligible, approximately: The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 8-bit in another scheme: mirai-l, where most use Q8_0.776.1 range771.4–779.688.8 range86.9–90.30.70 range0.70–0.7166,557 range65,117–71,3246of 6 planned
Magnitude 0.2.17 · Metal · Q8_0 Eligible, approximately: The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its phases were timed by the harness clock around the engine API, so they include transport that engine counters leave out.747.9 range731.2–756.844.4 range44.3–44.51.04 range1.04–1.0546,119 range46,034–46,2756of 6 planned
LiteRT-LM 0.19.1 · CPU · litert-int8 Eligible, approximately: The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 8-bit in another scheme: litert-int8, where most use Q8_0.198.9 range198.3–200.6635 range631–6431.25 range1.25–1.253,308 range3,268–3,3306of 6 planned
Rapid-MLX 1.0.2 · Metal · mlx-affine-8bit-g64 Eligible, approximately: The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 8-bit in another scheme: mlx-affine-8bit-g64, where most use Q8_0.712.2 range711.3–714.661.9 range59.5–65.32.05 range1.91–2.3533,458 range31,693–34,7906of 6 planned
MLX LM 0.32.2 · Metal · mlx-affine-8bit-g64 Eligible, approximately: The experiments use different effective KV cache precision: litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 8-bit in another scheme: mlx-affine-8bit-g64, where most use Q8_0.702.2 range695.5–703.157.7 range57.2–57.92.09 range1.94–2.3835,880 range35,780–36,1866of 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.

Decode speed tok/s, higher is better
Time to first token ms, lower is better
Peak memory GiB, lower is better
Prefill speed tok/s, higher is better

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-m5-max-64gb--macos-27--lfm2.5-350m--8-bit--archive-2048-128.