Skip to content

docs(binaries): archive the payload, not just the binary #307

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
glennneuber merged 3 commits into from
Sep 18, 2026
Merged

docs(binaries): archive the payload, not just the binary #307

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
8fc0ba2
docs(binaries): archive the payload, not just the binary
glennneuber Sep 18, 2026
9fc22fb
docs(binaries): 8a7ba949 is now serving on :11435
glennneuber Sep 18, 2026
8fa7034
docs(binaries): record golden parity at the d9add9d1 pin
glennneuber Sep 18, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
34 changes: 26 additions & 8 deletions docs/maxusai/vision-suite/BINARIES.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,8 @@ A host can be rebuilt from it; a binary sitting only in `/tmp` cannot.
| `0.32.14-maxusai-9594f81e` | [`9594f81e`](https://github.com/MaxusAI/ollama/commit/9594f81e) | llama.cpp **b10434** | v0.32.14 pin | `711d4ad126773ddf…` | the `mlx-metal-0-32-14` preflight baseline |
| `0.32.14-maxusai-c82b0464` | [`c82b0464`](https://github.com/MaxusAI/ollama/commit/c82b0464) | llama.cpp **b10434** | v0.32.14 pin | `03f9f9289dbaba1b…` | last pre-0.33.0 deploy on :11435 (2026-08-22 → 27); preflight PASS 2026-08-22; rollback target (Go half) for the 0.33.0 swap |
| `0.33.0-maxusai-21cfe88e` | [`21cfe88e`](https://github.com/MaxusAI/ollama/commit/21cfe88e) | llama.cpp **b10488** | 27fec909 pin | `6ab35025981be587…` | provenance for the `mlx-metal-0-33-0` profile measurement and the 27fec909 golden recalibration; rollback target (Go half) for the 0.33.2 swap |
| `0.33.2-maxusai-2b95b4a5` | [`2b95b4a5`](https://github.com/MaxusAI/ollama/commit/2b95b4a5) | llama.cpp **b10630** | c793734e pin | `cd6033d6e72a430e…` | current; the v0.33.2 fold (#232, ADR 0033 xgrammar). Provenance for the `mlx-metal-0-33-2` profile measurement; preflight PASS 2026-08-30, goldens NOT recalibrated at this MLX pin |
| `0.33.2-maxusai-2b95b4a5` | [`2b95b4a5`](https://github.com/MaxusAI/ollama/commit/2b95b4a5) | llama.cpp **b10630** | c793734e pin | `cd6033d6e72a430e…` | served :11435 2026-08-30 → 09-18; rollback target (BOTH halves archived); the v0.33.2 fold (#232, ADR 0033 xgrammar). Provenance for the `mlx-metal-0-33-2` profile measurement; preflight PASS 2026-08-30, goldens NOT recalibrated at this MLX pin |
| `0.34.0-maxusai-8a7ba949` | [`8a7ba949`](https://github.com/MaxusAI/ollama/commit/8a7ba949) | llama.cpp **b10864** | d9add9d1 pin | `867ee5d6063d8f77…` | the v0.34.0 + v0.34.1 folds (#297, #302). Provenance for the `mlx-metal-0-34-0` profile (#306); preflight PASS 2026-09-18 (19/12); whitespace bound (#301) confirmed behaviourally on real documents across the MLX pin move. **current** — deployed to `:11435` 2026-09-18 13:17, verified by version, a live MLX load reporting `0.32.2-61-gd9add9d`, and the dlopen path resolving the repo payload. Native gates pass: `x/mlxrunner` and all subpackages, and vision golden parity on 31b/26b/12b (max element delta 0.109 / 0.051 / 0.063) — the first build whose goldens were CHECKED at its MLX pin |

Full checksums:

Expand All @@ -28,6 +29,7 @@ d807360e94e0e17ac346df9bef198b6a182ef2f47bff78a0e772f6d1d67bad72 ~/.ollama/bina
03f9f9289dbaba1bba2a5826a22e1aa85525e96c7e78942bf41828ff90908a71 ~/.ollama/binaries/ollama-0.32.14-maxusai-c82b0464
6ab35025981be587ff0a73f0b7ae007b608300defb46f176065c8f3f52d78139 ~/.ollama/binaries/ollama-0.33.0-maxusai-21cfe88e
cd6033d6e72a430e0b96170adb3d7408c908edfdae392654faeaae46f31e9ee2 ~/.ollama/binaries/ollama-0.33.2-maxusai-2b95b4a5
867ee5d6063d8f77be6d8278c51d18f49892f00aa80c38c51630227aa9f68a57 ~/.ollama/binaries/ollama-0.34.0-maxusai-8a7ba949
```

## What b10353 is the provenance for
Expand Down Expand Up @@ -59,10 +61,19 @@ been rebuilt at MLX `c793734`, reported `MLX version=0.32.1-37-gc793734` — old
Go, new MLX, a pairing nothing was ever measured on. `mlx_payload_pin` failed it
against the profile's `27fec909`, which is what that check is for.

So "rollback target" below means **the Go half**. A true rollback restores the
payload too: check out the commit and rebuild (`CLEAN_DEPS=1`, below) so
`build/lib/ollama` holds that pin's MLX again, or verify what you actually got by
reading the engine-init line before trusting the restored server.
So "rollback target" meant **the Go half** only, and for the rows above
`0.33.2` it still does: a true rollback to those needs a checkout and rebuild
(`CLEAN_DEPS=1`, below) so `build/lib/ollama` holds that pin's MLX again.

**From 2026-09-18 the payload is archived too.** `~/.ollama/binaries/` now holds
`payload-0.33.2-maxusai-2b95b4a5/` and `payload-0.34.0-maxusai-8a7ba949/` beside
the binaries, so rolling between those two is a copy rather than a ~25 minute cold
rebuild — which on this host also depends on Xcode's separately-downloadable Metal
toolchain being installed. Archiving only the binary made every earlier row a
one-way door in practice; see "Adding one".

Either way, verify what you actually got by reading the engine-init line before
trusting a restored server.

## Rebuilding one

Expand All @@ -85,12 +96,19 @@ gates on.

## Adding one

Archive the binary whenever a payload pin moves, before the new build replaces
it on the host:
Archive **both halves** whenever a payload pin moves, before the new build
replaces them on the host. The binary alone is not enough — it does not carry
its MLX (above), so a binary-only archive cannot restore a working pairing:

```sh
cp -p /tmp/ollama-vs ~/.ollama/binaries/ollama-$(OLLAMA_HOST=127.0.0.1:1 /tmp/ollama-vs --version 2>&1 \
VER=$(OLLAMA_HOST=127.0.0.1:1 /tmp/ollama-vs --version 2>&1 \
| sed -n 's/^Warning: client version is //p')
cp -p /tmp/ollama-vs ~/.ollama/binaries/ollama-$VER
mkdir -p ~/.ollama/binaries/payload-$VER
cp -R build/lib/ollama/. ~/.ollama/binaries/payload-$VER/
```

Archive the OUTGOING pairing the same way before overwriting it, because
`build/lib/ollama` is the only copy of the payload the running server resolves.

Then add a row here with its sha256 and what it is the provenance for.
Loading