Comparison Not counted
Which engine runs LFM2.5 2.6B best on M5 Max · 64 GiB?
- Device
- M5 Max · 64 GiB (Apple silicon Macs)
- OS
- macOS 27
- Model
- LFM2.5 2.6B, 2.6B parameters
- Weights
- 4-bit
- Workload
- Archive 2048 input / 128 output: 2,048 tokens in, 128 out
- Decoding
- Greedy, batch 1, context 4,096
- Power
- Plugged in
Ranking
Memory: Peak memory while generating: the less, the better.
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 |
|---|---|---|---|---|---|---|
| LiteRT-LM 0.19.1 · CPU · litert-int4 | Eligible, approximately: The experiments use different effective KV cache precision: litert-lm default, f16, model-dtype. The experiments here use different 4-bit schemes: litert-int4, Q4_K_M, mlx-affine-4bit-g64. | 54.2 range52.4–54.6 | 10,200 range10,174–10,423 | 3.20 range3.20–3.21 | 201.7 range197.4–202.2 | 6of 6 planned |
| llama.cpp b12110 · Metal · Q4_K_M | Complete Its output check failed: 8/9 of the checks the model passes at BF16; needs 8 and the mandatory ones (needle-start, continuation-128). Left out, failed by every BF16 configuration: color, sort, needle-end. | 212.7 range208.7–217.0 | 313 range313–316 | 1.86 range1.86–1.86 | 6,555 range6,504–6,570 | 6of 6 planned |
| Magnitude 0.2.17 · Metal · Q4_K_M | Unsupported Magnitude's chat API renders LFM2.5 2.6B's own template, which always opens a thinking block, so it streams reasoning and can't run the closed-think prompts (measured in qualification). | – | – | – | – | 0of 6 planned |
| MLX LM 0.32.2 · Metal · mlx-affine-4bit-g64 | Complete Its output check failed: 7/9 of the checks the model passes at BF16; needs 8 and the mandatory ones (needle-start, continuation-128). Left out, failed by every BF16 configuration: color, sort, needle-end. | 258.4 range257.5–259.3 | 208 range208–211 | 2.98 range2.93–3.03 | 9,858 range9,759–9,883 | 6of 6 planned |
| Ollama 0.40.13 · Metal · Q4_K_M | Complete Its output check failed: 8/9 of the checks the model passes at BF16; needs 8 and the mandatory ones (needle-start, continuation-128). Left out, failed by every BF16 configuration: color, sort, needle-end. | 224.1 range223.7–225.4 | 297 range296–298 | 1.86 range1.86–1.86 | 6,966 range6,939–6,972 | 6of 6 planned |
| Rapid-MLX 1.0.2 · Metal · mlx-affine-4bit-g64 | Complete Its output check failed: 7/9 of the checks the model passes at BF16; needs 8 and the mandatory ones (needle-start, continuation-128). Left out, failed by every BF16 configuration: color, sort, needle-end. | 258.3 range257.5–259.7 | 234 range231–236 | 3.15 range3.10–3.20 | 8,790 range8,698–8,909 | 6of 6 planned |
| uzu 0.6.17 · Metal · mirai-m | Unsupported uzu can't run the frozen prompts: Error: uzu counted 23 input tokens, not the frozen prompt's 26 (measured in qualification). | – | – | – | – | 0of 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--lfm2.5-2.6b--4-bit--archive-2048-128.