jmigual commited on
Commit
2ffe61c
·
verified ·
1 Parent(s): 600c18f

Upload model

Browse files
README.md CHANGED
@@ -1,3 +1,45 @@
1
  ---
2
  license: apache-2.0
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: apache-2.0
3
+ base_model: Marqo/marqo-fashionCLIP
4
+ library_name: litert
5
+ tags:
6
+ - litert
7
+ - tflite
8
+ - fashion
9
+ - image-feature-extraction
10
  ---
11
+
12
+ # Marqo-FashionCLIP image encoder, LiteRT dynamic int8
13
+
14
+ The image tower of [Marqo/marqo-fashionCLIP](https://huggingface.co/Marqo/marqo-fashionCLIP)
15
+ (revision `44f4c655124ed71e90cbf528d82f508d69d8a81b`), converted to LiteRT and quantised. The
16
+ Robbeta Android app downloads it to compare outfit photos with the clothes in a wardrobe, on
17
+ the phone.
18
+
19
+ | File | Bytes | SHA-256 |
20
+ |---|---|---|
21
+ | `marqo-fashionclip-image-dynint8-v1.tflite` | 88 160 672 | `8aa00548e4f7768f97c9c11f1130d646bd6c1cf6b980451b5e9609a8295427b6` |
22
+
23
+ A later model gets a new file name; this file never changes.
24
+
25
+ ## Input and output
26
+
27
+ - Input `[1, 3, 224, 224]` float32, NCHW, RGB.
28
+ - Output `[1, 512]` float32, not normalised. Normalise it (L2) before taking cosines.
29
+ - Preprocessing, as open_clip does for this model: resize the shorter side to 224 with Pillow
30
+ bicubic, then centre-crop to 224 x 224 (Robbeta pads to a square first, so nothing is
31
+ cropped), scale to 0..1, subtract mean `(0.48145466, 0.4578275, 0.40821073)` and divide by std
32
+ `(0.26862954, 0.26130258, 0.27577711)`.
33
+
34
+ ## How it was made
35
+
36
+ `litert-torch` 0.9.4 converted `model.visual` from open_clip 3.3.0 (torch 2.13.0) to a float32
37
+ `.tflite`; `ai-edge-quantizer` 0.9.0 then applied the `dynamic_wi8_afp32` recipe (int8 weights,
38
+ float activations, dynamic-range int8 matmuls). The script is `tools/matcher/convert.py` in the
39
+ Robbeta repository.
40
+
41
+ ## Licence
42
+
43
+ Apache-2.0, the licence of the original model (see `LICENSE`). This file is a modified version
44
+ of Marqo-FashionCLIP's weights: converted to LiteRT and quantised to int8. Marqo-FashionCLIP is
45
+ by Marqo; see the original model card for its training data and evaluation.
marqo-fashionclip-image-dynint8-v1.tflite ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8aa00548e4f7768f97c9c11f1130d646bd6c1cf6b980451b5e9609a8295427b6
3
+ size 88160672