Regenerate 2026-09-24: 1005 recipes (Qwen-Image-2.1 on 21 cards)

Adds the Qwen-Image-2.1 24-32 GB and 8 GB tiers (6 recipes) and the 2026-09-24 refresh of all 21 Qwen-Image-2.1 pages. 100 models, 2700 compatibility rows, 166 benchmark rows (unchanged).

README.md

@@ -33,13 +33,13 @@ configs:
 
 Which open-weight AI models actually run on which consumer GPU, and what it takes
 to get them running. 2 700 model×GPU verdicts across 100 models
-and 27 cards, plus 999 full setup guides
+and 27 cards, plus 1 005 full setup guides
 (19 MB of markdown) written against specific hardware.
 
 This is the machine-readable form of [smeltcore.com](https://smeltcore.com). Every row carries a
 `url` back to the page it came from.
 
-Generated 2026-09-23T12:36:58+00:00 from the public read API (`https://api.smeltcore.com/api/v1`) — no private data, no
+Generated 2026-09-24T08:45:55+00:00 from the public read API (`https://api.smeltcore.com/api/v1`) — no private data, no
 credentials, reproducible by anyone.
 
 ## Configs
@@ -47,7 +47,7 @@ credentials, reproducible by anyone.
 | config | rows | what it is |
 |---|---:|---|
 | `compatibility` | 2 700 | **the point of the dataset.** One row per model × GPU, with a verdict |
-| `recipes` | 999 | full setup guides, markdown included, tagged by model / GPU / tool |
+| `recipes` | 1 005 | full setup guides, markdown included, tagged by model / GPU / tool |
 | `models` | 100 | the catalogue: licence, upstream repo, modality |
 | `gpus` | 27 | the cards: VRAM, vendor, series |
 | `benchmark_sources` | 166 | third-party measurements, normalised and cited |
@@ -65,8 +65,8 @@ The whole dataset turns on this column, so it is worth reading before using it.
 
 | verdict | rows | meaning |
 |---|---:|---|
-| `verified` | 998 | somebody ran it on this exact card and wrote down how — there is a recipe behind the row |
-| `fits` | 665 | inferred: the model's memory floor is under the card's VRAM, and the vendor is supported. Not measured |
+| `verified` | 1 004 | somebody ran it on this exact card and wrote down how — there is a recipe behind the row |
+| `fits` | 659 | inferred: the model's memory floor is under the card's VRAM, and the vendor is supported. Not measured |
 | `unknown` | 467 | no floor established for this model, so no honest call can be made |
 | `too_big` | 570 | the memory floor exceeds this card. This is the one verdict asserted from anywhere, not only from same-vendor evidence |
 
@@ -125,7 +125,7 @@ extend to the measurements themselves, which belong to the sites named in
   title  = {Self-Hosted AI — GPU Compatibility, Recipes and Catalogue},
   author = {smeltcore},
   url    = {https://smeltcore.com},
-  note   = {Generated 2026-09-23T12:36:58+00:00},
+  note   = {Generated 2026-09-24T08:45:55+00:00},
   year   = {2026}
 }
 ```

data/compatibility.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:989469402c8ff1d32af31c487b6d907e52c8a54469fcad976ff98098698fc078
-size 44346
+oid sha256:97e6dff78c069cfda07b7b2b6d466e8fc63372c9884f43b536163916b7e6142f
+size 44718

data/gpus.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:de824f45f658517d8bdac7902b3e3295b944ab2acf88e25c196658c47e7e96d2
-size 3354
+oid sha256:b1f5e6e6cb451853dbfdf2ca241cb9484e4c637db2b23763d81dd8d80b08f3cf
+size 3360

data/recipes.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8082337279564e15f271ee246ccd44500408be1f9257dcaa760ef62046dec935
-size 2985167
+oid sha256:11d7ef1a4415c5da08c29d5778bf0b1274cfa3816b485a2d22a5fcf3dae0f2ab
+size 3109495