From 0e2a4433a57e8922109675320367d6de31719782 Mon Sep 17 00:00:00 2001 From: Andrea Borio Date: Wed, 22 Jul 2026 16:24:26 +0200 Subject: [PATCH] docs: adopt canonical Hebrus model repositories --- docs/ENGINE_REFERENCE.md | 4 ++-- docs/HF_MODEL_RENAME_INVENTORY.md | 15 ++++++++++----- docs/REBRAND_ROLLOUT_PLAN.md | 11 ++++++----- docs/contracts/qwen-release.json | 2 +- docs/glm52-expert-major-v2.md | 2 +- docs/qwen-expert-major-store.md | 2 +- download_model.sh | 6 +++--- tests/qwen/README.md | 2 +- tests/test_download_model.sh | 4 ++-- tools/brand_boundary.json | 14 +++++++------- 10 files changed, 34 insertions(+), 28 deletions(-) diff --git a/docs/ENGINE_REFERENCE.md b/docs/ENGINE_REFERENCE.md index e2315b1c3..c5b6a700f 100644 --- a/docs/ENGINE_REFERENCE.md +++ b/docs/ENGINE_REFERENCE.md @@ -100,8 +100,8 @@ complete output SHA-256 in its publication record: output identity required by [`deepseek-expert-major-v2.md`](deepseek-expert-major-v2.md). The current canonical mirror is an offline converter source, not the runtime artifact. -- [GLM 5.2 DS4 GGUF](https://huggingface.co/andreaborio/GLM-5.2-DS4-GGUF) -- [Qwen3.6-35B-A3B DS4 GGUF](https://huggingface.co/andreaborio/Qwen3.6-35B-A3B-DS4-GGUF): +- [GLM 5.2 Hebrus GGUF](https://huggingface.co/andreaborio/GLM-5.2-Hebrus-GGUF) +- [Qwen3.6-35B-A3B Hebrus GGUF](https://huggingface.co/andreaborio/Qwen3.6-35B-A3B-Hebrus-GGUF): use only the MLX affine4/group-64 artifact identified in [`qwen-expert-major-store.md`](qwen-expert-major-store.md). The older Q4_K_S object is incompatible with this runtime. diff --git a/docs/HF_MODEL_RENAME_INVENTORY.md b/docs/HF_MODEL_RENAME_INVENTORY.md index 0a0228ba0..10656f619 100644 --- a/docs/HF_MODEL_RENAME_INVENTORY.md +++ b/docs/HF_MODEL_RENAME_INVENTORY.md @@ -1,16 +1,17 @@ # Hugging Face Model Rename Inventory -- Snapshot time: 2026-07-22, before any repository move +- Snapshot time: 2026-07-22, immediately before the repository moves +- Move status: all three in-place moves completed and verified on 2026-07-22 - Namespace: `andreaborio` - Visibility: all three repositories are public and ungated - Purpose: establish a reviewable baseline for the in-place DS4-to-Hebrus repository moves described in [`REBRAND_ROLLOUT_PLAN.md`](REBRAND_ROLLOUT_PLAN.md) -This snapshot is preparatory evidence, not authorization to move a repository. -Repeat the API inventory immediately before every move because revisions, -cards, counters, and file inventories can change after this document is -committed. +This snapshot is the pre-move baseline used for the completed operations. The +post-move comparison covered revisions, full sibling inventories, file sizes, +LFS SHA-256 values, visibility, gating, downloads, likes, Git heads, legacy URL +redirects, and ranged reads through each old repository ID. ## Repository baseline @@ -20,6 +21,10 @@ committed. | `andreaborio/GLM-5.2-DS4-GGUF` | `andreaborio/GLM-5.2-Hebrus-GGUF` | `0182407316ad8a3cf6cb676dad9ae298af8e091a` | 98 | 0 | 19 | | `andreaborio/Qwen3.6-35B-A3B-DS4-GGUF` | `andreaborio/Qwen3.6-35B-A3B-Hebrus-GGUF` | `7bf9c3f7f6136aeb2599d75ee61c0cc2f18e2b02` | 343 | 0 | 7 | +All old web and API endpoints returned HTTP `307` to the corresponding Hebrus +ID. Old pinned `resolve` requests redirected and completed as HTTP `206 Partial +Content`; old and new Git URLs returned the same recorded HEAD for every model. + ## Qualified runtime artifacts | Model | Immutable filename | Bytes | SHA-256 | diff --git a/docs/REBRAND_ROLLOUT_PLAN.md b/docs/REBRAND_ROLLOUT_PLAN.md index 809dfafe6..13035fc34 100644 --- a/docs/REBRAND_ROLLOUT_PLAN.md +++ b/docs/REBRAND_ROLLOUT_PLAN.md @@ -1,6 +1,6 @@ # Hebrus Rebrand Rollout Plan -- Status: active integration; compatibility PRs open, public renames gated +- Status: compatibility bridges merged; Hugging Face model moves completed - Last updated: 2026-07-22 - Scope: Hebrus engine, Hebrus Studio, the Hebrus Studio website, and the qualified Hugging Face model repositories @@ -27,10 +27,10 @@ stable naming boundary remains in | Surface | Canonical work | State | Immediate constraint | | --- | --- | --- | --- | -| Engine | `andreaborio/ds4`, PR [#6](https://github.com/andreaborio/ds4/pull/6) | Rebased on current `origin/main`; local complete premerge green; macOS CI green | Fix the Linux-only zsh runner gate, then require all CI and model-backed release lanes before release | -| Studio | `andreaborio/dsbox`, PR [#11](https://github.com/andreaborio/dsbox/pull/11) | Rebased on current `main`; 409 tests, build, audit, and local packaged-app verification green; dual HF IDs implemented | Require the rerun macOS package check and all PR checks to be green before merge | +| Engine | `andreaborio/ds4`, merged PR [#6](https://github.com/andreaborio/ds4/pull/6) at `7686f70` | Compatibility bridge merged; complete local premerge and Linux/macOS CI green | Model-backed release lanes remain required before a final release | +| Studio | `andreaborio/dsbox`, merged PR [#11](https://github.com/andreaborio/dsbox/pull/11) at `dd13cb5` | Compatibility bridge merged; 409 tests, build, audit, and local/CI packaged-app verification green | Publish and verify a post-merge development candidate before the final signed release | | Website | `andreaborio/hebrus-site`, `main` at `23507e2` | Published on GitHub Pages; clean install, zero-vulnerability audit, lint, static export, and Pages test green | Final logo, canonical post-rename links, and final signed download remain pending | -| Model Hub | Three consolidated `andreaborio/*-DS4-GGUF` repositories | Public pre-move snapshot recorded in [`HF_MODEL_RENAME_INVENTORY.md`](HF_MODEL_RENAME_INVENTORY.md); Studio bridge accepts future Hebrus IDs | Merge the compatibility bridge, repeat the inventory immediately before each authenticated move, then move one repository at a time | +| Model Hub | Three consolidated `andreaborio/*-Hebrus-GGUF` repositories | In-place moves completed; exact revisions, files, hashes, counters, Git heads, range downloads, and DS4-ID redirects verified | Refresh live model-card wording and retain continuous redirect/download monitoring | Release work must use the active engine rebrand branch, the active Studio bridge branch, and the GitHub Pages site checkout identified in the table above. A @@ -174,7 +174,8 @@ published artifact or historical attribution link was rewritten destructively. Hugging Face documents an in-place move operation that redirects the old URL and preserves download counts and likes. Use that operation rather than copying -or re-uploading weights. The initial public move set is: +or re-uploading weights. The initial public move set was completed on +2026-07-22: | Current repository | Future canonical repository | | --- | --- | diff --git a/docs/contracts/qwen-release.json b/docs/contracts/qwen-release.json index cd4ae3ad7..c6ff9eb44 100644 --- a/docs/contracts/qwen-release.json +++ b/docs/contracts/qwen-release.json @@ -2,7 +2,7 @@ "schemaVersion": 1, "modelFamily": "Qwen3.6-35B-A3B", "downloadTarget": "qwen-v2", - "repository": "andreaborio/Qwen3.6-35B-A3B-DS4-GGUF", + "repository": "andreaborio/Qwen3.6-35B-A3B-Hebrus-GGUF", "publishedArtifact": { "status": "published", "filename": "Qwen3.6-35B-A3B-DS4-ExpertMajor-v2-MLX-Affine4-G64.gguf", diff --git a/docs/glm52-expert-major-v2.md b/docs/glm52-expert-major-v2.md index ce9fbaafe..f6c179651 100644 --- a/docs/glm52-expert-major-v2.md +++ b/docs/glm52-expert-major-v2.md @@ -48,7 +48,7 @@ explicit 8K context is the predictable 64 GiB starting point; larger contexts should be admitted and measured independently. The artifact is hosted at -[`andreaborio/GLM-5.2-DS4-GGUF`](https://huggingface.co/andreaborio/GLM-5.2-DS4-GGUF). +[`andreaborio/GLM-5.2-Hebrus-GGUF`](https://huggingface.co/andreaborio/GLM-5.2-Hebrus-GGUF). | Property | Value | | --- | --- | diff --git a/docs/qwen-expert-major-store.md b/docs/qwen-expert-major-store.md index 3c9b161e9..13526c4e5 100644 --- a/docs/qwen-expert-major-store.md +++ b/docs/qwen-expert-major-store.md @@ -88,7 +88,7 @@ used for publication. | Item | Value | |---|---| | Publication state | `published` | -| Repository | `andreaborio/Qwen3.6-35B-A3B-DS4-GGUF` | +| Repository | `andreaborio/Qwen3.6-35B-A3B-Hebrus-GGUF` | | Artifact | `Qwen3.6-35B-A3B-DS4-ExpertMajor-v2-MLX-Affine4-G64.gguf` | | Artifact bytes | 20,808,566,880 | | Artifact SHA-256 | `dd17266185833a9f05531ce366fd7284ddca1ed64aa3dcf06e321e8c72c9ea3d` | diff --git a/download_model.sh b/download_model.sh index 5e1247eb7..e24c62687 100755 --- a/download_model.sh +++ b/download_model.sh @@ -1,16 +1,16 @@ #!/bin/sh set -eu -RUNTIME_DEEPSEEK_REPO="andreaborio/DeepSeek-V4-Flash-DS4-GGUF" +RUNTIME_DEEPSEEK_REPO="andreaborio/DeepSeek-V4-Flash-Hebrus-GGUF" RUNTIME_DEEPSEEK_FILE="DeepSeek-V4-Flash-IQ2XXS-w2Q2K-AProjQ8-SExpQ8-OutQ8-chat-v2-imatrix-DS4-ExpertMajor-v2.gguf" RUNTIME_DEEPSEEK_BYTES=86720114272 RUNTIME_DEEPSEEK_SHA256="8378080263eb9224f7228d72e2afa4ac3cf74a116023fdec2c596ff228a33e3f" -RUNTIME_GLM_REPO="andreaborio/GLM-5.2-DS4-GGUF" +RUNTIME_GLM_REPO="andreaborio/GLM-5.2-Hebrus-GGUF" RUNTIME_GLM_FILE="GLM-5.2-DS4-ExpertMajor-v2-Q2_K.gguf" RUNTIME_GLM_BYTES=262147193504 RUNTIME_GLM_SHA256="7f5017e3076e706c78f2a5322b035a9e2f6519c65ff5b6be8b2d91aeff61505d" RUNTIME_QWEN_STATUS="published" -RUNTIME_QWEN_REPO="andreaborio/Qwen3.6-35B-A3B-DS4-GGUF" +RUNTIME_QWEN_REPO="andreaborio/Qwen3.6-35B-A3B-Hebrus-GGUF" RUNTIME_QWEN_FILE="Qwen3.6-35B-A3B-DS4-ExpertMajor-v2-MLX-Affine4-G64.gguf" RUNTIME_QWEN_BYTES=20808566880 RUNTIME_QWEN_SHA256="dd17266185833a9f05531ce366fd7284ddca1ed64aa3dcf06e321e8c72c9ea3d" diff --git a/tests/qwen/README.md b/tests/qwen/README.md index 94e1debfd..f379159ea 100644 --- a/tests/qwen/README.md +++ b/tests/qwen/README.md @@ -114,7 +114,7 @@ Use this exact release identity for every Qwen model-backed gate: | Bytes | `20,808,566,880` | | SHA-256 | `dd17266185833a9f05531ce366fd7284ddca1ed64aa3dcf06e321e8c72c9ea3d` | | Routed storage | MLX affine4, group size 64 | -| Hugging Face | [`andreaborio/Qwen3.6-35B-A3B-DS4-GGUF`](https://huggingface.co/andreaborio/Qwen3.6-35B-A3B-DS4-GGUF) | +| Hugging Face | [`andreaborio/Qwen3.6-35B-A3B-Hebrus-GGUF`](https://huggingface.co/andreaborio/Qwen3.6-35B-A3B-Hebrus-GGUF) | The v2 file contains one routed-weight copy inside its checksummed embedded store. Canonical Qwen GGUFs are offline converter inputs, not inference diff --git a/tests/test_download_model.sh b/tests/test_download_model.sh index fcf2349bb..4b958abab 100755 --- a/tests/test_download_model.sh +++ b/tests/test_download_model.sh @@ -214,12 +214,12 @@ EOF run_runtime_success \ deepseek-v2 \ - andreaborio/DeepSeek-V4-Flash-DS4-GGUF \ + andreaborio/DeepSeek-V4-Flash-Hebrus-GGUF \ DeepSeek-V4-Flash-IQ2XXS-w2Q2K-AProjQ8-SExpQ8-OutQ8-chat-v2-imatrix-DS4-ExpertMajor-v2.gguf \ ds4-v0.2.0 run_runtime_success \ glm-v2 \ - andreaborio/GLM-5.2-DS4-GGUF \ + andreaborio/GLM-5.2-Hebrus-GGUF \ GLM-5.2-DS4-ExpertMajor-v2-Q2_K.gguf \ ds4-v0.2.0 run_runtime_success \ diff --git a/tools/brand_boundary.json b/tools/brand_boundary.json index 118539181..0913d0ac4 100644 --- a/tools/brand_boundary.json +++ b/tools/brand_boundary.json @@ -376,7 +376,7 @@ "location": "content", "token": "DS4", "classification": "migration-pending", - "maximum": 43 + "maximum": 39 }, { "path": "docs/HF_MODEL_RENAME_INVENTORY.md", @@ -698,7 +698,7 @@ "location": "content", "token": "DS4", "classification": "serialized/permanent", - "maximum": 3 + "maximum": 2 }, { "path": "docs/deepseek-expert-major-v2.md", @@ -740,7 +740,7 @@ "location": "content", "token": "DS4", "classification": "migration-pending", - "maximum": 11 + "maximum": 9 }, { "path": "docs/guides/MIGRATING_TO_HEBRUS.md", @@ -768,7 +768,7 @@ "location": "content", "token": "DS4", "classification": "migration-pending", - "maximum": 9 + "maximum": 8 }, { "path": "docs/releases/hebrus-bridge-launch-candidate.md", @@ -838,7 +838,7 @@ "location": "content", "token": "DS4", "classification": "migration-pending", - "maximum": 10 + "maximum": 7 }, { "path": "ds4.c", @@ -2217,7 +2217,7 @@ "location": "content", "token": "DS4", "classification": "migration-pending", - "maximum": 6 + "maximum": 4 }, { "path": "tests/qwen/UNICODE_DATA_PROVENANCE.md", @@ -2413,7 +2413,7 @@ "location": "content", "token": "DS4", "classification": "migration-pending", - "maximum": 12 + "maximum": 10 }, { "path": "tests/test_expert_major.py",