LFM2.5 2.6B on M5 Max · 64 GiB

Experiment Complete Published

llama.cpp b12110 on M5 Max · 64 GiB, Metal

LFM2.5 2.6B as Q8_0 (8-bit), archive 2048 input / 128 output: a 2048-token prompt and 128 tokens out. Measured 12 Nov 2026, 22:42 UTC by Glenn Sonna, Xybrid.

Eligible for ranking when the objective's required metrics are measured and this configuration is selected.Approximate comparison: The experiments use different effective KV cache precision: litert-lm default, f16, model-dtype. The experiments here use different 8-bit schemes: litert-int8, Q8_0, mlx-affine-8bit-g64.

Experiment id s3-lfm2.5-2.6b-8-bit-llama-2048

Summary

Over 6 valid trials of 6 attempted (6 planned), computed from the raw trials below.

MetricMedianRangeMean ± SDQuartiles
Decode speed tok/s, higher is better142.4 140.5–147.3 143.1 ± 2.8 140.8–144.9
Time to first token ms, lower is better298 297–299 298 ± 0.6 298–299
Peak memory GiB, lower is better2.98 2.98–2.98 2.98 ± 0.00 2.98–2.98
Prefill speed tok/s, higher is better6,885 6,874–6,908 6,886 ± 13.1 6,876–6,893

Trials, metric by metric

The line runs from the lowest to the highest valid trial, the dot is the median and each tick is one trial.

Every trial

Raw token counts and durations, with the speeds computed from them. Invalid trials stay in the record with their reason; they are left out of the summary, never silently dropped.

Every attempted trial
TrialTime to first token msPrefill tok/sDecode tok/sPeak memory GiBThermalNote
12996,8742,048 tok in 298 ms, engine141.3127 tok in 899 ms, engine2.98nominal → nominal
22986,8942,048 tok in 297 ms, engine140.5127 tok in 904 ms, engine2.98nominal → nominal
32976,9082,048 tok in 296 ms, engine143.5127 tok in 885 ms, engine2.98nominal → nominal
42996,8742,048 tok in 298 ms, engine147.3127 tok in 862 ms, engine2.98nominal → nominal
52996,8802,048 tok in 298 ms, engine145.4127 tok in 873 ms, engine2.98nominal → nominal
62986,8902,048 tok in 297 ms, engine140.6127 tok in 903 ms, engine2.98nominal → nominal

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.

Configuration

Device

Device
Apple Mac17,6
Cohort
Apple silicon Macs (desktop)
SoC
Apple M5 Max
CPU
18-core Apple CPU
GPU
Not recorded
NPU
Not recorded
RAM
64.00 GiB
OS
macOS 27.2.0 (26B123)
Kernel
27.2.0
Power
mains
Low power mode
No
Performance mode
Not recorded

Engine

Engine
llama.cpp
Maintainer
ggml-org
License
MIT, open source
Version
b12110
Commit
cd959bae9
Binary SHA-256
e7e3c60a5e612460d095bfd219047040d6bb9d0b56466aae30a480980b676959
Built from
release
Build flags
None
Backend
Metal
Threads
18
Interface
http-server
Settings in effect
context_length=4096 parallel=1 speculative=false backend=metal launch.log_verbosity=4 launch.log_colors=off launch.cache_prompt=false launch.cache_ram_mib=0 launch.cache_reuse=0 launch.continuous_batching=false launch.context_shift=false launch.flash_attn=on launch.cache_type_k=f16 launch.cache_type_v=f16 launch.gpu_layers=999 launch.load_mode=mmap offloaded_layers=31 launch.mmap=true n_batch=2048 n_ubatch=512 n_threads=18 n_threads_batch=18 flash_attn=enabled

Model and weights

Base model
LFM2.5 2.6B (LFM2.5, 2.6B)
Upstream
LiquidAI/LFM2.5-2.6B@7141b56cd4650af95c7c7b86983443aed050c96a
Weights
LiquidAI/LFM2.5-2.6B-GGUF@e7caca5d835a3901a8e0d63e94009429bafafdfc
Format
gguf
File 1
LFM2.5-2.6B-Q8_0.gguf · 2.68 GiB · sha256 1e22128dfa128bdfb684da167e74e072d0a056baa7d06d9f280291e2839b0fc9
File 2
README.md · 3,068 bytes · sha256 d08b79260b3998e96792777085e4175b730667568f3a953ad4ac9099cfa27056
Quantization
Q8_0 (8-bit)
Weight bits
8
Group size
32
KV cache
f16
Conversion
Published upstream by LiquidAI
Conversion command
Not recorded

Workload

Workload
Archive 2048 input / 128 output
Prompt
2,048 tokens, from prompts/source.txt
Prompt SHA-256
c10d1b7bef62d95322f8f8d229dc66118d9375107bdc49b06f76287d058b2f6c
Chat template
ea663864491de7ade391839479860ca95541f892f72665c73251fbd4643b1bef
Output
128 tokens, ignoring end of sequence
Context length
4,096
Batch size
1
Sampling
Greedy
Timing
One streamed request per trial
Warmup
2 runs, not counted
Trials planned
6
Cooldown
15 s between trials

Conditions

Ambient
Not recorded
Thermal at start
nominal
Thermal at end
nominal
Thermal gate
macOS thermal state nominal
Throttling detected
No

Provenance

Measured by
Glenn Sonna, Xybrid
Operator affiliation
Xybrid
Run at
12 Nov 2026, 22:42 UTC
Harness
xybrid-bench 0.4.0 (777cb3d9fecaa7cc1a30694d98492c33dbc6d693)
Methodology
v0.6
Publication
published
Published
16 Nov 2026, 10:23 UTC
Reviewed by
Glenn Sonna

Output check

Result
Passed
Method
sanity-v1-bf16-reference
Reference
checks/sanity-v1.json
Reference SHA-256
a702dee079525ae0ec90465b1c9e26052c99cb3e1c53e83dea05be4c194fc7ca
Required passing checks
8
Check arithmetic
Passed · sha256 68ce25cb2c4b530f85434cd352428b022149c641163300e566f4f9ee9d789bcf
Check capital
Passed · sha256 b1098d691672b55e98767ac84c6d8375f7bcc0d78154c861ea70ac2ff66e0ba7
Check opposite
Passed · sha256 7c90b06eef25984adf77ec226311829bbc1065a033e9ad6b26d9bd0dfdc65f95
Check weekday
Passed · sha256 03db656ee680fb1053be9c98631722488bb10269075159dbae5b000f0359dae2
Check color
Failed · sha256 a4849893c59e4aa15ff645cadd4cce0fbd882097352a75d8e6f23619578a73c6
Check sort
Failed · sha256 6904619ddb316a20bc1220a9088da9a4181f1f16766b6fdd188defd9cddc4c75
Check animal
Passed · sha256 3e9114b6df29561433c18b3a6abce2c635fbd8e9ff25b5fcee93149bfea52859
Check word
Passed · sha256 d1bf31457072aaa4c116ea5819a644456232fdd87b15203d3eafdd5e541f4e61
Check copy
Passed · sha256 d8c711dfa5fef68d1d0f86320daeca065b91bc8d0dbcea210f8b9ef80eefc6c2
Check needle-start
Passed · sha256 56a69645d0ff053b3b7001c224f07a374bf963a22c9fa4706dda5e56e3616870
Check needle-end
Failed · sha256 0f97f855a97e7b397b129f50bdc8840f19221695dc7199f2c62ea9875b504298
Check continuation-128
Passed · sha256 35b435d7f1c75bb05c07d58cad000525a1d62d6dbe7fd00ae7416953dfa26286
Detail
9/9 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.

Reproduce it

Commands
pnpm bench run --plan plans/s3-lfm2.5-2.6b-8-bit.json

Three palindromic blocks per quantization class and workload; load, two warmups, one trial, unload per turn.

Raw data