M725 commited on
Commit
a47ce37
Β·
verified Β·
1 Parent(s): 72a2b11

upload cortexa-create-scorer v1

Browse files
README.md ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - en
4
+ license: other
5
+ license_name: pleius-internal
6
+ tags:
7
+ - onnx
8
+ - video-classification
9
+ - short-form-video
10
+ - creator-tools
11
+ ---
12
+
13
+ # cortexa-create-scorer (v10-proxy)
14
+
15
+ Production scorer for Pleius Create (short-form video). Uses the same
16
+ weights as `M725/cortexa-marketing-scorer` but with a different pillar
17
+ remap suited to short-form video.
18
+
19
+ ## Pipeline
20
+
21
+ ```
22
+ video ─→ ffmpeg 1fps sample β†’ N frames
23
+ N frames ─→ CLIP ViT-B/32 vision encoder ─→ (N, 512) stim_emb
24
+ "general audience" ─→ CLIP text β†’ demo_emb
25
+ cortexa_v10 head per frame β†’ per-frame curves (engagement, audience_drive, ...)
26
+ ↓
27
+ aggregate to 4 Create pillars:
28
+ hook = mean engagement, first up-to-3 frames
29
+ hold = mean engagement, all frames
30
+ algorithmic_fit = mean audience_drive
31
+ brand_lift = mean demographic_appeal
32
+ overall (win) = sigmoid(audience_drive_mean Β· 6)
33
+ ```
34
+
35
+ The remap mirrors `hf-space/app/content.py` in the serving Space
36
+ `M725/cortexa-ad-api`. When a dedicated v64a_PROD weight ships, this
37
+ repo will be replaced with that ranker; until then v10-proxy is the
38
+ production behavior.
39
+
40
+ ## Files
41
+
42
+ Same 4 ONNX files as `M725/cortexa-marketing-scorer` (they are the same
43
+ weights, served under a different surface and pillar mapping). See that
44
+ repo's README for file-level details.
45
+
46
+ ## License
47
+
48
+ Pleius internal β€” see https://pleius.com. Not for redistribution.
clip_text_fp16.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cfde6cbc76dd5784a0da8201710c480a32ebef88d46da38cb0eec8ee9a8af60a
3
+ size 127400687
clip_tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
clip_vision_fp16.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:90648a4ed39e5b79083084be70f56e3389e44b3e8bcfadd1280341e1ffb252b5
3
+ size 180800618
cortexa_v10.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:add9910a147d66867a779746d33f5c883de05c685c403206e553dcbd5f8c63f6
3
+ size 3731886