Repin Parakeet Redux to the current upstream revision

moondream/parakeet-redux was rewritten upstream on 2026-09-23 to a single
commit (2bf128600aac), so the pinned revision fad622f no longer resolves.
model.safetensors still downloads from the old revision, but config.json,
ternary.json, tokenizer.json and README.md return HTTP 404, which makes
`install.sh --install-model` fail.

The weights, config, ternary map and tokenizer at 2bf1286 have the same
size and SHA-256 as the old pins, so the model itself is unchanged. Only
the model card README.md differs (8533 -> 8832 bytes), so its pin is
updated along with the revision.

Fixes #1

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/build.md

@@ -93,7 +93,7 @@ and target application compatibility require further headset work.
 
 ## Inference runtime
 
-Redux weights at `fad622f25f303105c20d70e201bcc477c88b620c` are CC-BY-4.0. Pinned
+Redux weights at `2bf128600aac4b16946f7ed8372e56117fe5e23b` are CC-BY-4.0. Pinned
 file sizes/hashes and attribution live in `assets/backends/redux.json`; offline
 `--list-models`/`--check-model redux --model-dir /absolute/model` (or
 `scripts/model-status.py`) verify without inference or downloads. Manifest

docs/design.md

@@ -226,9 +226,12 @@ may deliver at most once; cancellation invalidates it before any later reply.
 ## On-device Redux
 
 Use the **same** `moondream/parakeet-redux` revision from the benchmark:
-`fad622f25f303105c20d70e201bcc477c88b620c` (177,774,490-byte weight file), initially
+`2bf128600aac4b16946f7ed8372e56117fe5e23b` (177,774,490-byte weight file), initially
 moondream 2.4.0 / kestrel 0.8.0. Do not substitute dense Ultra or silently fall
-back to desktop/cloud inference.
+back to desktop/cloud inference. The benchmark ran on revision
+`fad622f25f303105c20d70e201bcc477c88b620c`, which no longer resolves after the
+upstream history was rewritten; the weights, config, ternary map and tokenizer at
+`2bf1286` are byte-identical to it (same SHA-256), only the model card README differs.
 
 The earlier benchmark used this local API (its source belongs to the originating
 repository, not this project):

docs/packaging.md

@@ -45,7 +45,7 @@ prefix. ARM64/glibc packaging is not a claim of compatibility with arbitrary Lin
 ```sh
 python3 scripts/package-release.py --stage /path/to/stage --output /existing/output \
   --version 0.1.202609241627 --arch linux-aarch64 \
-  --model-revision fad622f25f303105c20d70e201bcc477c88b620c --external-runtime
+  --model-revision 2bf128600aac4b16946f7ed8372e56117fe5e23b --external-runtime
 ```
 
 Use the actual binary's numeric `MAJOR.MINOR.YYYYMMDDHHMM` UTC version

docs/third-party.md

@@ -27,7 +27,7 @@ source-only, and Python packages are fetched from PyPI on the user's machine.
 
 | Component | Where | License |
 | --- | --- | --- |
-| Parakeet Redux weights, revision `fad622f25f303105c20d70e201bcc477c88b620c`, derived from NVIDIA Parakeet TDT 0.6B v3 by Moondream/M87 Labs; pinned sizes and hashes in `assets/backends/redux.json` | <https://huggingface.co/moondream/parakeet-redux> | CC-BY-4.0 |
+| Parakeet Redux weights, revision `2bf128600aac4b16946f7ed8372e56117fe5e23b`, derived from NVIDIA Parakeet TDT 0.6B v3 by Moondream/M87 Labs; pinned sizes and hashes in `assets/backends/redux.json` | <https://huggingface.co/moondream/parakeet-redux> | CC-BY-4.0 |
 | Kestrel, `kestrel-kernels` | <https://github.com/m87-labs/kestrel> | Local inference is free and needs no API key, per the Kestrel README. Finetuned-model inference needs an API key and is not used here. |
 | `moondream` | PyPI | See the package |
 | PyTorch (CPU build) | <https://github.com/pytorch/pytorch> | BSD-3-Clause |

docs/worker.md

@@ -57,7 +57,7 @@ device="cpu", cpu_threads=threads)` and persistent
 Install an **isolated** Python runtime with the
 moondream 2.4.0, kestrel 0.8.0 and compatible CPU dependencies; provide
 preinstalled local weights from revision
-`fad622f25f303105c20d70e201bcc477c88b620c` and pass its directory
+`2bf128600aac4b16946f7ed8372e56117fe5e23b` and pass its directory
 explicitly. The code verifies exact sizes and SHA-256 of weights/config/tokenizer
 against `assets/backends/redux.json` via the shared, offline
 `python/frameyap/model_files.py` schema/verifier before importing model libraries.