Comparison Counted
Which engine runs LFM2.5 1.2B Instruct best on M5 Max · 64 GiB?
- Device
- M5 Max · 64 GiB (Apple silicon Macs)
- OS
- macOS 27
- Model
- LFM2.5 1.2B Instruct, 1.2B 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
Ranking
Speed: Decode speed: tokens generated per second.
| # | Engine | Decode and First token |
|---|---|---|
| 1 | uzu†, an approximate comparison, see details | Decode: 540.7 tok/s 75.2 ms first token |
| 2 | ExecuTorch†, an approximate comparison, see details | Decode: 472.1 tok/s 52.3 ms first token |
| 3 | Ollama†, an approximate comparison, see details | Decode: 454.6 tok/s 30.6 ms first token |
| 4 | Magnitude†, an approximate comparison, see details | Decode: 444.5 tok/s 26.6 ms first token |
| 5 | llama.cpp†, an approximate comparison, see details | Decode: 438.1 tok/s 34.2 ms first token |
| Not ranked in this view | ||
| Not ranked | LiteRT-LM | Not counted Its output check failed: 10/12 of the checks the model passes at BF16; needs 11 and the mandatory ones (needle-start, needle-end, continuation-128). Left out, failed by every BF16 configuration: none. |
| Not ranked | MLX LM | Not counted Its output check failed: 10/12 of the checks the model passes at BF16; needs 11 and the mandatory ones (needle-start, needle-end, continuation-128). Left out, failed by every BF16 configuration: none. |
| Not ranked | Rapid-MLX | Not counted Its output check failed: 10/12 of the checks the model passes at BF16; needs 11 and the mandatory ones (needle-start, needle-end, continuation-128). Left out, failed by every BF16 configuration: none. |
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: bf16, fp32, litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 4-bit in another scheme: mirai-m, where most use Q4_K_M.
† ExecuTorch: approximate comparison. The experiments use different effective KV cache precision: bf16, fp32, litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 4-bit in another scheme: mlx-4w-g64, where most use Q4_K_M.
† Ollama: approximate comparison. The experiments use different effective KV cache precision: bf16, fp32, litert-lm default, f16, magnitude default, model-dtype, uzu default.
† Magnitude: approximate comparison. The experiments use different effective KV cache precision: bf16, fp32, 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.
† llama.cpp: approximate comparison. The experiments use different effective KV cache precision: bf16, fp32, litert-lm default, f16, magnitude default, model-dtype, uzu default.
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.
| Experiment | Ranking eligibility | Decode tok/s, higher is better | First token ms, lower is better | Memory GiB, lower is better | Prefill tok/s, higher is better | Valid trials |
|---|---|---|---|---|---|---|
| uzu 0.6.17 · Metal · mirai-m | Eligible, approximately: The experiments use different effective KV cache precision: bf16, fp32, litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 4-bit in another scheme: mirai-m, where most use Q4_K_M. | 540.7 range539.8–541.4 | 75.2 range74.9–76.9 | 0.90 range0.90–0.96 | 13,205 range13,070–13,700 | 6of 6 planned |
| ExecuTorch 1.6.0 · Metal · mlx-4w-g64 | Eligible, approximately: The experiments use different effective KV cache precision: bf16, fp32, litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 4-bit in another scheme: mlx-4w-g64, where most use Q4_K_M. | 472.1 range470.4–475.7 | 52.3 range51.6–52.5 | Not measured | 4,923 range4,830–5,020 | 6of 6 planned |
| Ollama 0.40.13 · Metal · Q4_K_M | Eligible, approximately: The experiments use different effective KV cache precision: bf16, fp32, litert-lm default, f16, magnitude default, model-dtype, uzu default. | 454.6 range452.3–458.8 | 30.6 range23.9–33.6 | 0.93 range0.93–0.95 | 8,803 range7,994–11,423 | 6of 6 planned |
| Magnitude 0.2.17 · Metal · Q4_K_M | Eligible, approximately: The experiments use different effective KV cache precision: bf16, fp32, 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. | 444.5 range437.1–446.0 | 26.6 range26.2–26.8 | 1.56 range1.55–1.56 | 9,627 range9,540–9,771 | 6of 6 planned |
| llama.cpp b12110 · Metal · Q4_K_M | Eligible, approximately: The experiments use different effective KV cache precision: bf16, fp32, litert-lm default, f16, magnitude default, model-dtype, uzu default. | 438.1 range430.6–443.7 | 34.2 range33.5–34.9 | 0.93 range0.93–0.95 | 7,671 range7,514–7,899 | 6of 6 planned |
| ExecuTorch 1.6.0 · CPU · 8da4w | Eligible, approximately: The experiments use different effective KV cache precision: bf16, fp32, litert-lm default, f16, magnitude default, model-dtype, uzu default. Its weights are 4-bit in another scheme: 8da4w, where most use Q4_K_M. | 158.8 range156.6–161.8 | 262 range261–272 | Not measured | 977.1 range941.2–980.8 | 6of 6 planned |
| LiteRT-LM 0.19.1 · CPU · litert-int4 | Complete Its output check failed: 10/12 of the checks the model passes at BF16; needs 11 and the mandatory ones (needle-start, needle-end, continuation-128). Left out, failed by every BF16 configuration: none. | 103.5 range102.7–103.8 | 716 range715–724 | 1.23 range1.23–1.23 | 370.5 range366.9–371.3 | 6of 6 planned |
| LiteRT-LM 0.19.1 · GPU · litert-int4-gpu | Complete Its output check failed: 8/12 of the checks the model passes at BF16; needs 11 and the mandatory ones (needle-start, needle-end, continuation-128). Left out, failed by every BF16 configuration: none. | 330.8 range327.6–340.2 | 114 range104–117 | 1.82 range1.82–1.83 | 3,438 range3,380–3,497 | 6of 6 planned |
| MLX LM 0.32.2 · Metal · mlx-affine-4bit-g64 | Complete Its output check failed: 10/12 of the checks the model passes at BF16; needs 11 and the mandatory ones (needle-start, needle-end, continuation-128). Left out, failed by every BF16 configuration: none. | 582.9 range574.9–589.7 | 28.9 range28.6–29.3 | 0.89 range0.87–0.90 | 9,024 range8,897–9,092 | 6of 6 planned |
| Rapid-MLX 1.0.2 · Metal · mlx-affine-4bit-g64 | Complete Its output check failed: 10/12 of the checks the model passes at BF16; needs 11 and the mandatory ones (needle-start, needle-end, continuation-128). Left out, failed by every BF16 configuration: none. | 575.8 range523.4–582.9 | 34.8 range34.2–36.6 | 1.12 range1.11–1.14 | 7,493 range7,106–7,622 | 6of 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-m5-max-64gb--macos-27--lfm2.5-1.2b-instruct--4-bit--archive-256-128.