Bonsai 27B field guide
How to verify the exact Ternary Bonsai 27B model
Record the Hugging Face revision, file size, hash, and runtime name before comparing Bonsai 27B results.
How do you verify which Ternary Bonsai 27B checkpoint you tested?
Record the Hugging Face repository, immutable revision, exact weight file size, and SHA 256 hash. Our trace used prism-ml/Ternary-Bonsai-27B-mlx-2bit at revision 70f75f3ad081ab840a42f3304c02c27e7f89bfb7. The model.safetensors file was 8,490,785,104 bytes with SHA 256 8acd4597893ea7004e2d7336c3cf6e3157b8896592bbcf066db004021e45846b.
Why the model name is not enough
A repository can change while keeping the same name. A runtime can also give the model a shorter local identifier. The revision and hash connect a result to one exact file.
Format also matters. The MLX bundle, GGUF file, ideal ternary size, and unpacked FP16 checkpoint are different artifacts.
What to save with every result
- Repository and revision.
- File name, byte count, and hash.
- Runtime and version.
- Hardware, memory, context, and prompt shape.
Questions people ask
Does a local LM Studio model ID prove the checkpoint?
No. It is a local name. Save the source repository and file hash too.
Why publish the full hash?
It lets another tester confirm that both machines used the same bytes.
Primary sources
Vendor claims on this page are labeled as PrismML claims. Local results are added only after a saved trace completes.