Ternary Bonsai 2 27B on M5 Max · 64 GiB

Experiment Complete Published

MLX LM 0.32.2 on M5 Max · 64 GiB, Metal

Ternary Bonsai 2 27B as mlx-affine-2bit-g128 (2-bit), archive 256 input / 128 output: a 256-token prompt and 128 tokens out. Measured 9 Nov 2026, 23:07 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: f16, model-dtype. Its weights are 2-bit in another scheme: mlx-affine-2bit-g128, where most use Q2_0.

Experiment id s2-bonsai2-27b-2-bit-mlx-256

Summary

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

MetricMedianRangeMean ± SDQuartiles
Decode speed tok/s, higher is better45.3 44.4–45.3 45.2 ± 0.4 45.2–45.3
Time to first token ms, lower is better421 419–493 433 ± 29.3 420–425
Peak memory GiB, lower is better9.65 9.64–9.65 9.65 ± 0.00 9.64–9.65
Prefill speed tok/s, higher is better609.2 520.6–612.7 594.6 ± 36.4 604.6–611.8

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
1493520.6256 tok in 492 ms, engine44.4127 tok in 2,858 ms, engine9.64nominal → nominal
2419612.7256 tok in 418 ms, engine45.3127 tok in 2,801 ms, engine9.65nominal → nominal
3425603.8256 tok in 424 ms, engine45.2127 tok in 2,809 ms, engine9.64nominal → nominal
4420611.9256 tok in 418 ms, engine45.3127 tok in 2,801 ms, engine9.65nominal → nominal
5423607.1256 tok in 422 ms, engine45.3127 tok in 2,805 ms, engine9.65nominal → nominal
6420611.2256 tok in 419 ms, engine45.3127 tok in 2,804 ms, engine9.65nominal → 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
MLX LM
Maintainer
Apple (ml-explore)
License
MIT, open source
Version
0.32.2
Commit
Not recorded
Binary SHA-256
9d46803799f99d6fc7c35fff2d4357c79b1a25ae98cc63efeaaebeb50a7b5f33
Built from
package
Build flags
None
Backend
Metal
Threads
Not recorded
Interface
python-worker
Settings in effect
backend=Device(gpu, 0) prompt_cache=false speculative=false greedy=true temperature=0 sampler=argmax prefill_step_size=2048 context_length=4096 kv_quantized=false batch_size=1 trust_remote_code=false weight_bits=2 weight_group_size=128

Model and weights

Base model
Ternary Bonsai 2 27B (Bonsai, 27B)
Upstream
prism-ml/Ternary-Bonsai-2-27B-gguf@839bd57023c1558945afea70a74fa83c054f35f9
Weights
prism-ml/Ternary-Bonsai-2-27B-mlx-2bit@24b5937a4a104a44f790b6ccbd0c79b6fabd4299
Format
safetensors
File 1
NOTICE.txt · 411 bytes · sha256 de0e0c48fb6f691a31e74f338e3ccf93f9ecdfe2866ab769c4bf8b79a7636a30
File 2
README.md · 23,372 bytes · sha256 99abaa471b920c6a77c4a32441ecc14e13c148e23318fffe5e1a4d6169831650
File 3
chat_template.jinja · 8,952 bytes · sha256 c3cf9e34abf4f9e36c2d72165aa9c132d3e2a725b6c2586aaa3a8af9d7a81041
File 4
config.json · 58,145 bytes · sha256 238de7c512cc56a733421e3fd011d88f8260739e3d00e32c5d65b7943cc9f837
File 5
files.json · 3,122 bytes · sha256 f0d3b0108d8e52685b81da7916e9943ff9cf2f470fbcc6b24f66a3d562492f1c
File 6
generation_config.json · 202 bytes · sha256 e70c136c1b78ddc1fb0905bac8e733a4dc448d4f852a5dd75143fffc70be550e
File 7
hadamard.json · 297,903 bytes · sha256 7132a3ec364f0bdac1f08f905f24f0ad2f14245060f592637a0396826d3b5fe6
File 8
model.safetensors · 8.01 GiB · sha256 130de5925082c168b7866b2e91b52e44abbafc99017e3ca352b77b5b55a269ed
File 9
preprocessor_config.json · 390 bytes · sha256 27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516
File 10
reload-validation.json · 86 bytes · sha256 41ec215fe19fa8ad59527fca6594a782d1b7f23b69948ef063311c768217788e
File 11
tokenizer-validation.json · 63 bytes · sha256 b36f33d09bd3de3e19de3fd2cf493a69ec0a5e24acf648b0c9fba120d9357554
File 12
tokenizer.json · 12.2 MiB · sha256 0997f410c57a1f4e53b09e4be8f4a172d90edd9564368fb0847030937229b9f3
File 13
tokenizer_config.json · 17,928 bytes · sha256 b11349aafa7cdc6a320767cf7ceb29ed82f7eda5d65e8e0819e76f0ce947bf27
File 14
video_preprocessor_config.json · 385 bytes · sha256 7768af27c1fafa9cc9011c1dc20067e03f8915e03b63504550e11d5066986d13
Quantization
mlx-affine-2bit-g128 (2-bit)
Weight bits
2
Group size
128
KV cache
model-dtype
Conversion
Published upstream by prism-ml
Conversion command
Not recorded

Workload

Workload
Archive 256 input / 128 output
Prompt
256 tokens, from prompts/source.txt
Prompt SHA-256
c10d1b7bef62d95322f8f8d229dc66118d9375107bdc49b06f76287d058b2f6c
Chat template
c3cf9e34abf4f9e36c2d72165aa9c132d3e2a725b6c2586aaa3a8af9d7a81041
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
9 Nov 2026, 23:07 UTC
Harness
xybrid-bench 0.4.0 (103beb8cf54e7e006744c19f3189370cd245047e)
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
11
Check arithmetic
Passed · sha256 6b51d431df5d7f141cbececcf79edf3dd861c3b4069f0b11661a3eefacbba918
Check capital
Passed · sha256 5dd272b4f316b776a7b8e3d0894b37e1e42be3d5d3b204b8a5836cc50597a6b1
Check opposite
Passed · sha256 e272deecde720ffc773d7a6a81adabbc366ccbb64da5abdb6b26f52cd3f952cb
Check weekday
Passed · sha256 7d8af1de1262f150187d5938bf649d6c35726970685f916d1ec6392a801b9762
Check color
Passed · sha256 d486dfbd5fb578340bccbdd0a662527eab38550648d5f44517e5ac71b8824703
Check sort
Passed · sha256 a386b3c9b4b4786df5bb6474bab0a62b8476e2f3f9c8a6433aca40152840f6b7
Check animal
Passed · sha256 4b227777d4dd1fc61c6f884f48641d02b4d121d3fd328cb08b5531fcacdabf8a
Check word
Passed · sha256 3b635d3b4437e023e72a9df26201b6ab51285b376446446ebb379121994550f6
Check copy
Passed · sha256 d61e6890d0e3571b0c7693612d41f8d059a1fcdef4ba959c0af4144541d41907
Check needle-start
Passed · sha256 8b9af2e5b8b4cfddb87d171997b95c471f8d68a68e30fb5177abeb3332372edd
Check needle-end
Passed · sha256 e6589c736af32f81bf239c5eda6de41959122ccc1d05463b75e0eb0c70fc831c
Check continuation-128
Passed · sha256 ed26731fe8d39dacac77703998aa4bb784317ae565cb21a58f83f75d7e458142
Detail
12/12; needs 11 and every mandatory check (needle-start, needle-end, continuation-128); the model has no BF16 reference.

Reproduce it

Commands
pnpm bench run --plan plans/s2-bonsai2-27b-2-bit.json

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

Raw data