arraypress commited on
Commit
3e8b112
·
verified ·
1 Parent(s): 9895af6

SigLIP 2 base/16 224 for Core AI (pix)

Browse files
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ pix-siglip2-base-224.aimodel/main.mlirb filter=lfs diff=lfs merge=lfs -text
LICENSE ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ Apache License
3
+ Version 2.0, January 2004
4
+ http://www.apache.org/licenses/
5
+
6
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
7
+
8
+ 1. Definitions.
9
+
10
+ "License" shall mean the terms and conditions for use, reproduction,
11
+ and distribution as defined by Sections 1 through 9 of this document.
12
+
13
+ "Licensor" shall mean the copyright owner or entity authorized by
14
+ the copyright owner that is granting the License.
15
+
16
+ "Legal Entity" shall mean the union of the acting entity and all
17
+ other entities that control, are controlled by, or are under common
18
+ control with that entity. For the purposes of this definition,
19
+ "control" means (i) the power, direct or indirect, to cause the
20
+ direction or management of such entity, whether by contract or
21
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
22
+ outstanding shares, or (iii) beneficial ownership of such entity.
23
+
24
+ "You" (or "Your") shall mean an individual or Legal Entity
25
+ exercising permissions granted by this License.
26
+
27
+ "Source" form shall mean the preferred form for making modifications,
28
+ including but not limited to software source code, documentation
29
+ source, and configuration files.
30
+
31
+ "Object" form shall mean any form resulting from mechanical
32
+ transformation or translation of a Source form, including but
33
+ not limited to compiled object code, generated documentation,
34
+ and conversions to other media types.
35
+
36
+ "Work" shall mean the work of authorship, whether in Source or
37
+ Object form, made available under the License, as indicated by a
38
+ copyright notice that is included in or attached to the work
39
+ (an example is provided in the Appendix below).
40
+
41
+ "Derivative Works" shall mean any work, whether in Source or Object
42
+ form, that is based on (or derived from) the Work and for which the
43
+ editorial revisions, annotations, elaborations, or other modifications
44
+ represent, as a whole, an original work of authorship. For the purposes
45
+ of this License, Derivative Works shall not include works that remain
46
+ separable from, or merely link (or bind by name) to the interfaces of,
47
+ the Work and Derivative Works thereof.
48
+
49
+ "Contribution" shall mean any work of authorship, including
50
+ the original version of the Work and any modifications or additions
51
+ to that Work or Derivative Works thereof, that is intentionally
52
+ submitted to Licensor for inclusion in the Work by the copyright owner
53
+ or by an individual or Legal Entity authorized to submit on behalf of
54
+ the copyright owner. For the purposes of this definition, "submitted"
55
+ means any form of electronic, verbal, or written communication sent
56
+ to the Licensor or its representatives, including but not limited to
57
+ communication on electronic mailing lists, source code control systems,
58
+ and issue tracking systems that are managed by, or on behalf of, the
59
+ Licensor for the purpose of discussing and improving the Work, but
60
+ excluding communication that is conspicuously marked or otherwise
61
+ designated in writing by the copyright owner as "Not a Contribution."
62
+
63
+ "Contributor" shall mean Licensor and any individual or Legal Entity
64
+ on behalf of whom a Contribution has been received by Licensor and
65
+ subsequently incorporated within the Work.
66
+
67
+ 2. Grant of Copyright License. Subject to the terms and conditions of
68
+ this License, each Contributor hereby grants to You a perpetual,
69
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
70
+ copyright license to reproduce, prepare Derivative Works of,
71
+ publicly display, publicly perform, sublicense, and distribute the
72
+ Work and such Derivative Works in Source or Object form.
73
+
74
+ 3. Grant of Patent License. Subject to the terms and conditions of
75
+ this License, each Contributor hereby grants to You a perpetual,
76
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
77
+ (except as stated in this section) patent license to make, have made,
78
+ use, offer to sell, sell, import, and otherwise transfer the Work,
79
+ where such license applies only to those patent claims licensable
80
+ by such Contributor that are necessarily infringed by their
81
+ Contribution(s) alone or by combination of their Contribution(s)
82
+ with the Work to which such Contribution(s) was submitted. If You
83
+ institute patent litigation against any entity (including a
84
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
85
+ or a Contribution incorporated within the Work constitutes direct
86
+ or contributory patent infringement, then any patent licenses
87
+ granted to You under this License for that Work shall terminate
88
+ as of the date such litigation is filed.
89
+
90
+ 4. Redistribution. You may reproduce and distribute copies of the
91
+ Work or Derivative Works thereof in any medium, with or without
92
+ modifications, and in Source or Object form, provided that You
93
+ meet the following conditions:
94
+
95
+ (a) You must give any other recipients of the Work or
96
+ Derivative Works a copy of this License; and
97
+
98
+ (b) You must cause any modified files to carry prominent notices
99
+ stating that You changed the files; and
100
+
101
+ (c) You must retain, in the Source form of any Derivative Works
102
+ that You distribute, all copyright, patent, trademark, and
103
+ attribution notices from the Source form of the Work,
104
+ excluding those notices that do not pertain to any part of
105
+ the Derivative Works; and
106
+
107
+ (d) If the Work includes a "NOTICE" text file as part of its
108
+ distribution, then any Derivative Works that You distribute must
109
+ include a readable copy of the attribution notices contained
110
+ within such NOTICE file, excluding those notices that do not
111
+ pertain to any part of the Derivative Works, in at least one
112
+ of the following places: within a NOTICE text file distributed
113
+ as part of the Derivative Works; within the Source form or
114
+ documentation, if provided along with the Derivative Works; or,
115
+ within a display generated by the Derivative Works, if and
116
+ wherever such third-party notices normally appear. The contents
117
+ of the NOTICE file are for informational purposes only and
118
+ do not modify the License. You may add Your own attribution
119
+ notices within Derivative Works that You distribute, alongside
120
+ or as an addendum to the NOTICE text from the Work, provided
121
+ that such additional attribution notices cannot be construed
122
+ as modifying the License.
123
+
124
+ You may add Your own copyright statement to Your modifications and
125
+ may provide additional or different license terms and conditions
126
+ for use, reproduction, or distribution of Your modifications, or
127
+ for any such Derivative Works as a whole, provided Your use,
128
+ reproduction, and distribution of the Work otherwise complies with
129
+ the conditions stated in this License.
130
+
131
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
132
+ any Contribution intentionally submitted for inclusion in the Work
133
+ by You to the Licensor shall be under the terms and conditions of
134
+ this License, without any additional terms or conditions.
135
+ Notwithstanding the above, nothing herein shall supersede or modify
136
+ the terms of any separate license agreement you may have executed
137
+ with Licensor regarding such Contributions.
138
+
139
+ 6. Trademarks. This License does not grant permission to use the trade
140
+ names, trademarks, service marks, or product names of the Licensor,
141
+ except as required for reasonable and customary use in describing the
142
+ origin of the Work and reproducing the content of the NOTICE file.
143
+
144
+ 7. Disclaimer of Warranty. Unless required by applicable law or
145
+ agreed to in writing, Licensor provides the Work (and each
146
+ Contributor provides its Contributions) on an "AS IS" BASIS,
147
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
148
+ implied, including, without limitation, any warranties or conditions
149
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
150
+ PARTICULAR PURPOSE. You are solely responsible for determining the
151
+ appropriateness of using or redistributing the Work and assume any
152
+ risks associated with Your exercise of permissions under this License.
153
+
154
+ 8. Limitation of Liability. In no event and under no legal theory,
155
+ whether in tort (including negligence), contract, or otherwise,
156
+ unless required by applicable law (such as deliberate and grossly
157
+ negligent acts) or agreed to in writing, shall any Contributor be
158
+ liable to You for damages, including any direct, indirect, special,
159
+ incidental, or consequential damages of any character arising as a
160
+ result of this License or out of the use or inability to use the
161
+ Work (including but not limited to damages for loss of goodwill,
162
+ work stoppage, computer failure or malfunction, or any and all
163
+ other commercial damages or losses), even if such Contributor
164
+ has been advised of the possibility of such damages.
165
+
166
+ 9. Accepting Warranty or Additional Liability. While redistributing
167
+ the Work or Derivative Works thereof, You may choose to offer,
168
+ and charge a fee for, acceptance of support, warranty, indemnity,
169
+ or other liability obligations and/or rights consistent with this
170
+ License. However, in accepting such obligations, You may act only
171
+ on Your own behalf and on Your sole responsibility, not on behalf
172
+ of any other Contributor, and only if You agree to indemnify,
173
+ defend, and hold each Contributor harmless for any liability
174
+ incurred by, or claims asserted against, such Contributor by reason
175
+ of your accepting any such warranty or additional liability.
176
+
177
+ END OF TERMS AND CONDITIONS
178
+
179
+ APPENDIX: How to apply the Apache License to your work.
180
+
181
+ To apply the Apache License to your work, attach the following
182
+ boilerplate notice, with the fields enclosed by brackets "[]"
183
+ replaced with your own identifying information. (Don't include
184
+ the brackets!) The text should be enclosed in the appropriate
185
+ comment syntax for the file format. We also recommend that a
186
+ file or class name and description of purpose be included on the
187
+ same "printed page" as the copyright notice for easier
188
+ identification within third-party archives.
189
+
190
+ Copyright [yyyy] [name of copyright owner]
191
+
192
+ Licensed under the Apache License, Version 2.0 (the "License");
193
+ you may not use this file except in compliance with the License.
194
+ You may obtain a copy of the License at
195
+
196
+ http://www.apache.org/licenses/LICENSE-2.0
197
+
198
+ Unless required by applicable law or agreed to in writing, software
199
+ distributed under the License is distributed on an "AS IS" BASIS,
200
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
201
+ See the License for the specific language governing permissions and
202
+ limitations under the License.
NOTICE ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ pix-siglip2-base-224
2
+ Copyright 2025 Google LLC (SigLIP 2 model weights and tokenizer)
3
+
4
+ This repository redistributes Google's SigLIP 2 base/16 at 224
5
+ (https://huggingface.co/google/siglip2-base-patch16-224, commit
6
+ 75de2d55ec2d0b4efc50b3e9ad70dba96a7b2fa2), licensed under the Apache License,
7
+ Version 2.0 (see LICENSE).
8
+
9
+ Modifications (made for swift-photo-index, https://github.com/arraypress/swift-photo-index):
10
+ - The vision and text towers were exported, unchanged in architecture, to an
11
+ Apple Core AI asset (pix-siglip2-base-224.aimodel) with coreai-torch 0.4.3,
12
+ weights cast from float32 to float16, as two entry points (image, text) that
13
+ return the pooled embeddings (get_image_features / get_text_features).
14
+ - The Gemma tokenizer from the same commit was rewritten into two flat binary
15
+ files (pix-siglip2-support/vocab.bin, merges.bin) with the same ids and
16
+ merge order.
17
+ The export script is included as export_siglip2.py.
README.md ADDED
@@ -0,0 +1,79 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: google/siglip2-base-patch16-224
4
+ tags:
5
+ - vision
6
+ - siglip2
7
+ - image-text-retrieval
8
+ - core-ai
9
+ - apple-silicon
10
+ - macos
11
+ library_name: swift-photo-index
12
+ ---
13
+
14
+ # pix-siglip2-base-224: SigLIP 2 for Core AI (macOS 27)
15
+
16
+ Google's [SigLIP 2 base/16 at 224](https://huggingface.co/google/siglip2-base-patch16-224)
17
+ (Tschannen et al., *SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic
18
+ Understanding, Localization, and Dense Features*, 2025; Apache-2.0) exported to an Apple Core AI
19
+ `.aimodel`. It puts pictures and sentences in one 768-dimensional space, so a photo of a dog on a
20
+ beach lands near the words "dog on a beach".
21
+
22
+ This is what meaning search in [`pix`](https://github.com/arraypress/swift-pix-cli) runs on, through
23
+ [swift-photo-index](https://github.com/arraypress/swift-photo-index). `pix index` stores one vector per
24
+ image (one per video: the mean of its sampled frames). `pix "dog on a beach"` embeds the query and
25
+ ranks by cosine. Words that are literally there (text read off the picture, speech, the file name)
26
+ still rank first, so an exact invoice number beats anything that only matches by meaning. pix
27
+ downloads this repository by itself the first time it needs it, at a pinned commit, and checks every
28
+ file against its published hash. Everything runs on device.
29
+
30
+ ## Files
31
+
32
+ | Path | What |
33
+ |---|---|
34
+ | `pix-siglip2-base-224.aimodel/` | Two entry points. `image`: `pixel_values` `[1,3,224,224]` float32 → `embedding` `[1,768]`. `text`: `ids` `[1,64]` int32 → `embedding` `[1,768]`. Float16 weights, 750 MB. |
35
+ | `pix-siglip2-support/vocab.bin`, `merges.bin` | The checkpoint's Gemma tokenizer (256,000 tokens) as flat binaries: `u32` count, then per id `u32` length + UTF-8; merges as `u32` (left, right, merged) id triples in rank order. |
36
+ | `pix-siglip2-support/config.json` | Source model and revision, sizes, normalisation, special-token ids. |
37
+ | `export_siglip2.py` | The export script (also writes the parity fixtures). |
38
+
39
+ The host does the rest, as `transformers` does. Images: resize to 224×224 without keeping the
40
+ aspect ratio, RGB, `(x/255 − 0.5)/0.5`, channel-first. Text: lower-case, spaces to `▁`, byte-fallback
41
+ BPE, append `<eos>`=1, right-pad with `<pad>`=0 to 64. Then L2-normalise both outputs and take
42
+ dot products.
43
+
44
+ ## Fidelity
45
+
46
+ Measured from Swift through Core AI against `transformers` (float32) on the same inputs. Tokenizer
47
+ ids are identical on 10 test strings (accents, emoji, CJK, Arabic, tabs, truncation at 64). Text
48
+ embeddings: worst cosine 0.9999993. Image embeddings: cosine 0.99999 when no resize is needed, and
49
+ 0.997 on a 400×300 image, where Core Graphics' resampling differs slightly from PIL's.
50
+
51
+ ## Retrieval
52
+
53
+ 1,000 COCO val2017 images (the first 1,000 by id) against their 5,003 captions, embedded through
54
+ this asset in Swift:
55
+
56
+ - Text → image recall: R@1 **0.720**, R@5 **0.916**, R@10 **0.959**.
57
+ - Through pix's own search (meaning blended with words, 0.09 cut-off): R@1 0.719, R@5 0.911,
58
+ R@10 0.948. Words alone find none of the captions (0/0/0), because full sentences never appear
59
+ in OCR text or labels.
60
+ - Cosine of true pairs: 5th percentile 0.101, median 0.150. Wrong pairs: 99th percentile 0.087,
61
+ 99.9th 0.135. pix's 0.09 floor keeps 97.5% of true pairs and drops 99.1% of wrong ones.
62
+ - Speed (Apple silicon, GPU): 12 ms per image, 4 ms per sentence, 2 s cold model load. Scoring a
63
+ query against 21,000 stored float16 vectors takes about 25–35 ms.
64
+
65
+ ## Use
66
+
67
+ ```sh
68
+ pix index ~/Pictures # fetches this model on first use
69
+ pix "a red car at the beach at sunset"
70
+ ```
71
+
72
+ To install it by hand, run `hf download arraypress/pix-siglip2-base-224 --local-dir models` and then
73
+ `pix model install models`. Requires macOS 27 (Core AI) on Apple silicon. Converted with coreai-torch
74
+ 0.4.3, coreai-core 1.0.0b3, torch 2.9.0 and transformers 5.12.1.
75
+
76
+ ## Licence
77
+
78
+ Apache-2.0, the same as the original. The weights and tokenizer are Google's; see `NOTICE` for what
79
+ was changed.
export_siglip2.py ADDED
@@ -0,0 +1,129 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # /// script
2
+ # requires-python = ">=3.11,<3.13"
3
+ # dependencies = ["torch==2.9.0", "numpy", "pillow", "coreai-torch==0.4.3", "coreai-core==1.0.0b3", "transformers==5.12.1"]
4
+ # ///
5
+ """SigLIP 2 base/16 at 224 (Google, Apache-2.0) → pix-siglip2-base-224.aimodel, its tokenizer and fixtures.
6
+
7
+ uv run Tools/export_siglip2.py [--out Tools/exports] [--dtype float16]
8
+
9
+ Two entry points in one asset:
10
+ image: pixel_values [1,3,224,224] float32, already resized and normalised ((x/255 - 0.5)/0.5) → embedding [1,768]
11
+ text: ids [1,64] int32 (lower-cased, Gemma BPE, <eos>=1 appended, right-padded with <pad>=0) → embedding [1,768]
12
+ Both are the model's pooled outputs (what get_image_features / get_text_features return); the host
13
+ normalises to unit length. The tokenizer is written as two flat binary files the Swift side reads
14
+ without a JSON parse: the vocabulary and the merges as id triples in rank order.
15
+ Fixtures: texts with their ids and normalised embeddings, and two synthetic images with theirs, so
16
+ the Swift tokenizer, preprocessing and asset are each checked against transformers."""
17
+ import argparse, json, shutil, struct, time
18
+ from pathlib import Path
19
+ import numpy as np, torch, torch.nn as nn
20
+ from PIL import Image
21
+ from transformers import AutoModel, AutoProcessor, AutoTokenizer
22
+ import coreai_torch
23
+ from coreai.runtime import AIModelAssetMetadata
24
+
25
+ HERE = Path(__file__).resolve().parent
26
+ ap = argparse.ArgumentParser()
27
+ ap.add_argument("--name", default="google/siglip2-base-patch16-224")
28
+ ap.add_argument("--revision", default="75de2d55ec2d0b4efc50b3e9ad70dba96a7b2fa2")
29
+ ap.add_argument("--asset", default="pix-siglip2-base-224.aimodel")
30
+ ap.add_argument("--dtype", default="float16", choices=["float16", "float32"])
31
+ ap.add_argument("--out", default=str(HERE / "exports"))
32
+ ap.add_argument("--fixtures", default=str(HERE.parent / "Tests/PhotoIndexTests/Fixtures/siglip"))
33
+ args = ap.parse_args()
34
+ LENGTH, SIZE = 64, 224
35
+
36
+ model = AutoModel.from_pretrained(args.name, revision=args.revision, dtype=torch.float32).eval()
37
+ proc = AutoProcessor.from_pretrained(args.name, revision=args.revision)
38
+ tok = AutoTokenizer.from_pretrained(args.name, revision=args.revision)
39
+
40
+ class Image_(nn.Module):
41
+ def __init__(self, m): super().__init__(); self.v = m.vision_model
42
+ def forward(self, pixel_values): return self.v(pixel_values=pixel_values).pooler_output
43
+ class Text_(nn.Module):
44
+ def __init__(self, m): super().__init__(); self.t = m.text_model
45
+ def forward(self, ids): return self.t(input_ids=ids).pooler_output
46
+ image_net, text_net = Image_(model).eval(), Text_(model).eval()
47
+
48
+ def ids_for(texts):
49
+ return tok([t.lower() for t in texts], padding="max_length", max_length=LENGTH, truncation=True, return_tensors="pt")["input_ids"].to(torch.int32)
50
+
51
+ out = Path(args.out); out.mkdir(parents=True, exist_ok=True)
52
+ support = out / "pix-siglip2-support"; support.mkdir(exist_ok=True)
53
+ fx = Path(args.fixtures); fx.mkdir(parents=True, exist_ok=True)
54
+
55
+ # Tokenizer → vocab.bin (u32 count, then u32 length + UTF-8 bytes per id) and merges.bin (u32 count, then u32 left, right, merged per rank).
56
+ tj = json.loads(tok.backend_tokenizer.to_str())
57
+ vocab = tj["model"]["vocab"]; inv = sorted(vocab.items(), key=lambda kv: kv[1])
58
+ assert [i for _, i in inv] == list(range(len(inv)))
59
+ with open(support / "vocab.bin", "wb") as f:
60
+ f.write(struct.pack("<I", len(inv)))
61
+ for s, _ in inv:
62
+ b = s.encode("utf-8"); f.write(struct.pack("<I", len(b))); f.write(b)
63
+ merges = tj["model"]["merges"]
64
+ with open(support / "merges.bin", "wb") as f:
65
+ f.write(struct.pack("<I", len(merges)))
66
+ for m in merges:
67
+ a, b = (m if isinstance(m, list) else m.split(" ", 1))
68
+ f.write(struct.pack("<III", vocab[a], vocab[b], vocab[a + b]))
69
+ config = {"model": args.name, "revision": args.revision, "dimension": model.config.vision_config.hidden_size,
70
+ "image_size": SIZE, "mean": 0.5, "std": 0.5, "text_length": LENGTH, "pad": tok.pad_token_id, "eos": tok.eos_token_id,
71
+ "unk": tok.unk_token_id, "lowercase": True, "space": "▁", "entry_points": {"image": "pixel_values", "text": "ids"}}
72
+ json.dump(config, open(support / "config.json", "w"), indent=1, ensure_ascii=False)
73
+ shutil.copy(support / "config.json", fx / "config.json")
74
+ print(f"tokenizer: {len(inv)} tokens, {len(merges)} merges", flush=True)
75
+
76
+ # Fixtures.
77
+ texts = ["a dog running on the beach", "The red car at the beach at sunset", "ERROR 517 EXCLUSIVITY VIOLATION",
78
+ " leading and trailing spaces ", "two spaces\tand a tab\nand a newline", "café résumé naïve — punctuation!!",
79
+ "emoji 🎹🎛️ and CJK 漢字 and Arabic عربى", "x", "1234567890 numbers 3.14",
80
+ " ".join(f"word{i}" for i in range(80))]
81
+ ids = ids_for(texts)
82
+ with torch.no_grad():
83
+ temb = nn.functional.normalize(text_net(ids.long()), dim=-1)
84
+ ref = nn.functional.normalize(model.get_text_features(input_ids=ids.long()).pooler_output if hasattr(model.get_text_features(input_ids=ids.long()), "pooler_output") else model.get_text_features(input_ids=ids.long()), dim=-1)
85
+ print(f"text wrapper vs get_text_features max |diff| {(temb - ref).abs().max():.2e}", flush=True)
86
+ temb.numpy().astype(np.float32).tofile(fx / "text_embeddings.f32")
87
+ json.dump({"texts": texts, "ids": ids.tolist()}, open(fx / "text_fixture.json", "w"), ensure_ascii=False)
88
+
89
+ ys, xs = np.mgrid[0:SIZE, 0:SIZE]
90
+ square = np.stack([xs * 255 // (SIZE - 1), ys * 255 // (SIZE - 1), (255 - xs * 255 // (SIZE - 1))], -1).astype(np.uint8)
91
+ Image.fromarray(square).save(fx / "gradient-224.png")
92
+ yw, xw = np.mgrid[0:300, 0:400]
93
+ wide = np.stack([((xw // 40 + yw // 40) % 2) * 200 + 30, xw * 255 // 399, yw * 255 // 299], -1).astype(np.uint8)
94
+ Image.fromarray(wide).save(fx / "checker-400x300.png")
95
+ with torch.no_grad():
96
+ embs = []
97
+ for name in ["gradient-224.png", "checker-400x300.png"]:
98
+ px = proc(images=Image.open(fx / name).convert("RGB"), return_tensors="pt")["pixel_values"]
99
+ embs.append(nn.functional.normalize(image_net(px), dim=-1)[0])
100
+ torch.stack(embs).numpy().astype(np.float32).tofile(fx / "image_embeddings.f32")
101
+ px = proc(images=Image.open(fx / "gradient-224.png").convert("RGB"), return_tensors="pt")["pixel_values"]
102
+ manual = torch.from_numpy(((square.astype(np.float32) / 255 - 0.5) / 0.5).transpose(2, 0, 1)[None])
103
+ print(f"processor vs manual normalise at 224 (no resize): max |diff| {(px - manual).abs().max():.2e}", flush=True)
104
+
105
+ # Export.
106
+ dtype = torch.float16 if args.dtype == "float16" else torch.float32
107
+ t0 = time.time()
108
+ conv = coreai_torch.TorchConverter(mode=coreai_torch.TorchConverter.Mode.RELEASE)
109
+ inet, tnet = Image_(model.to(dtype)).eval(), Text_(model.to(dtype)).eval()
110
+ class Cast(nn.Module):
111
+ def __init__(self, inner, castin): super().__init__(); self.inner, self.castin = inner, castin
112
+ def forward(self, x): return self.inner(x.to(dtype) if self.castin else x).to(torch.float32)
113
+ with torch.no_grad():
114
+ ex = torch.export.export(Cast(inet, True), (torch.zeros(1, 3, SIZE, SIZE),)).run_decompositions(coreai_torch.get_decomp_table())
115
+ conv.add_exported_program(ex, input_names=["pixel_values"], output_names=["embedding"], entrypoint_name="image")
116
+ ex = torch.export.export(Cast(tnet, False), (torch.ones(1, LENGTH, dtype=torch.int32),)).run_decompositions(coreai_torch.get_decomp_table())
117
+ conv.add_exported_program(ex, input_names=["ids"], output_names=["embedding"], entrypoint_name="text")
118
+ program = conv.to_coreai()
119
+ meta = AIModelAssetMetadata()
120
+ meta.author = "Google — SigLIP 2 base/16 224; Core AI export by swift-photo-index"
121
+ meta.license = "Apache-2.0"
122
+ meta.model_description = (f"SigLIP 2 ({args.name}@{args.revision[:7]}), {args.dtype}. Entry 'image': pixel_values [1,3,224,224] float32 "
123
+ "((x/255-0.5)/0.5, RGB, resized to 224x224) → embedding [1,768]. Entry 'text': ids [1,64] int32 (lower-cased Gemma BPE, "
124
+ "<eos>=1 appended, right-padded with 0) → embedding [1,768]. Normalise both in the host; cosine ranks.")
125
+ meta.creation_date = int(time.time())
126
+ asset = out / args.asset
127
+ if asset.exists(): shutil.rmtree(asset)
128
+ program.save_asset(asset, meta)
129
+ print(f"saved {asset} ({sum(f.stat().st_size for f in asset.rglob('*'))/1e6:.0f} MB) in {time.time()-t0:.0f} s", flush=True)
pix-siglip2-base-224.aimodel/main.hash ADDED
@@ -0,0 +1 @@
 
 
1
+ |N+ E��/�oi� ��>��,n�w
pix-siglip2-base-224.aimodel/main.mlirb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7c4e2b0c45e6c42fcf1e1f6f69c10bc2c13ec4ec17102c6e11e9770df09a9b35
3
+ size 750505678
pix-siglip2-base-224.aimodel/metadata.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "creationDate" : "20261013T183249Z",
3
+ "author" : "Google — SigLIP 2 base\/16 224; Core AI export by swift-photo-index",
4
+ "license" : "Apache-2.0",
5
+ "assetVersion" : "2.0",
6
+ "description" : "SigLIP 2 (google\/siglip2-base-patch16-224@75de2d5), float16. Entry 'image': pixel_values [1,3,224,224] float32 ((x\/255-0.5)\/0.5, RGB, resized to 224x224) → embedding [1,768]. Entry 'text': ids [1,64] int32 (lower-cased Gemma BPE, <eos>=1 appended, right-padded with 0) → embedding [1,768]. Normalise both in the host; cosine ranks.",
7
+ "producer" : "coreai-core 1.0.0b3"
8
+ }
pix-siglip2-support/config.json ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "google/siglip2-base-patch16-224",
3
+ "revision": "75de2d55ec2d0b4efc50b3e9ad70dba96a7b2fa2",
4
+ "dimension": 768,
5
+ "image_size": 224,
6
+ "mean": 0.5,
7
+ "std": 0.5,
8
+ "text_length": 64,
9
+ "pad": 0,
10
+ "eos": 1,
11
+ "unk": 3,
12
+ "lowercase": true,
13
+ "space": "▁",
14
+ "entry_points": {
15
+ "image": "pixel_values",
16
+ "text": "ids"
17
+ }
18
+ }
pix-siglip2-support/merges.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:219b2dbc1e645054ef8adacb4ed96a70d27b7b38e24fa848818c424bdf477f69
3
+ size 6967252
pix-siglip2-support/vocab.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:22aa6308f573f129d84205a72132a5f369247b7e4cd6bcd51dc4115d2be83a86
3
+ size 2955906