Ternary-Bonsai-27B · file← Ternary-Bonsai-27B
MLX 2-bit, first copy
An MLX copy already on the Mac Studio in August. Deleted on 7 September, never tested. Deleted 2026-09-07.
- Size
- 7.9 GB
- Status
- Deleted 7 Sep 2026
- Scored runs
- 0
ResultsNever scored
This file never completed a scored run. Its record below says why.
The recordHow it was made · how it was served · what went wrong · what happened to it
Rebuilt from the lab's records by Claude, an AI model. The results above come straight from the automated runs. How these pages are made.
| File | ternary-bonsai-27b-mlx/ (7.9 GB) |
| Format | MLX 2-bit |
| Where from | held by 2026-08-23; source not on record |
| Status | deleted 2026-09-07 17:36Z, no RESTORE entry |
| Hugging Face | https://huggingface.co/prism-ml/Ternary-Bonsai-27B-mlx-2bit (inferred) |
Where it came from
- What it was:
models/ternary-bonsai-27b-mlx, 7.9 GB, an MLX 2-bit pack of Ternary-Bonsai-27B. A lab note the assistant read on 2026-08-23 lists it as "ternary-bonsai-27b-mlx(2-bit, 22.0 t/s at 7.9 GB)". The source repo, revision, download date and who downloaded it are not on record; the 22.0 tok/s figure and the line itself came from that note. - Probably the same weights as Ternary-Bonsai-27B MLX 2-bit (prism-ml 0 ) (HF
mainsha 70f75f3 has been unchanged since 2026-07-14) — (reconstructed), nothing compares the two. - Related, not this file: a separate
bonsai-image-ternary-4bdirectory (3.6G, an image model) was deleted in the same command.
How it was served
| label | dates (UTC) | binary and version | ctx / -np / KV / -fa / --jinja / --reasoning-format | harness profile | why this label exists |
|---|---|---|---|---|---|
| none | no run recorded; on 2026-08-23 its path was the worked example for the model field of an MLX server request (a filesystem path) | mlx_lm server | n/a | n/a | No run-log label. A 2026-08-27 audit of MLX directories listed it with no suite baseline (-- NONE --). |
Seats held: none.
Decisions
- Why it existed: not on record. Alan, 2026-09-10: "Let’s bench it, really valid bench, I dabbled with it before", which may refer to this copy or to the July M4 Pro benchmark of the 8B (a lab note); (reconstructed) he had a Bonsai on the machine before the programme measured one.
- Deleted on 2026-09-07 at Alan's go-ahead. The assistant's triage listed
ternary-bonsai-27b-mlx(7.9G) as dead weight with no preset and no test and later summarised that tier as "never tested, no preset"; Alan: "Go for it and what’s the blocker on Xcode? I can be installed from cli". The 17:36Z command removed it (witholmoe,olmoe-colibriand the image model) viarm -rf.
What went wrong
- Deleted before it was ever benched, and no
RESTORE-deleted-*.shentry records it (checked 2026-10-23: nothing inRESTORE-deleted-20260908.sh,-20260909.sh,-20260911.shor the 2026-10-11 scripts names this directory under the 09-07 deletion). It was re-downloaded on 2026-09-14 for a proper bench (Ternary-Bonsai-27B MLX 2-bit (prism-ml 4 )). - Standing checks: no runs, so no partial runs, stale stamps or missing flags.
What happened to it
Held by 2026-08-23 → listed in the 2026-09-05 reclaimable audit as a 7.9 GB directory with neither a preset nor a run-log entry → deleted 2026-09-07 17:36Z. Re-downloadable from prism-ml/Ternary-Bonsai-27B-mlx-2bit.