upload cortexa-marketing-feedback v1

README.md

@@ -0,0 +1,66 @@
+---
+language:
+- en
+license: other
+license_name: pleius-internal
+tags:
+- onnx
+- conditional-text-generation
+- ad-feedback
+- distillation
+- creator-tools
+---
+
+# cortexa-marketing-feedback (distilled student)
+
+A ~4.4M-parameter conditional decoder distilled from
+`M725/cortexa-marketing-scorer` outputs. Takes CLIP-ViT-B/32 vision
+features (768-d) + the 4 Marketing pillar scores (or a "no-scores"
+sentinel for fast mode) and emits a creator-vernacular phrase chain:
+
+```
+"scroll stopping | clear cta | thumb stopping"
+"forgettable | looks clean | low contrast text"
+"lazy design | model looks fake | low contrast"
+```
+
+The student is meant to be the *feedback callout* shown on the result
+screen for paid users — plain-language pros and cons that go alongside
+the scorer's numeric output.
+
+## Files
+
+| file | purpose |
+|---|---|
+| `student_int8.onnx` | TinyTransformer decoder, 4 layers / 256-dim / 4 heads, INT8 dynamic-quantized. 6.9 MB. |
+| `tokenizer.json`    | Whole-phrase tokenizer (vocab ~115; specials `<pad>`, `<bos>`, `<eos>`, `<sep>`). |
+| `config.json`       | Encoder dim, pillar names, vocab size, special-token ids — read by the TS/JS runtime to shape inputs. |
+
+## Inference shape
+
+```
+inputs:
+  encoder_feats   (1, 768)  float32   # mean-pooled CLIP-ViT-B/32 vision output
+  scores          (1, 4)    float32   # [universal_appeal, demographic_appeal, audience_drive, engagement] in [0,1]
+  scores_present  (1,)      float32   # 1.0 anchored, 0.0 fast-mode
+  input_ids       (1, T)    int64     # decoder context
+outputs:
+  logits          (1, T, V) float32
+```
+
+Greedy decode works; **temperature 0.8 + top-k 20 + SEP-veto** is the
+recommended sampling config when running on more than one input
+(prevents the greedy "forgettable | forgettable | forgettable" collapse
+the v0 model exhibited).
+
+## Training
+
+15k phrase triples from 5k COCO photos. Each photo scored locally
+against the cortexa_v10 head; phrase chains generated by
+`research.distill_adjectives.phrase_rules.scores_to_phrase`. 12 epochs,
+AdamW, cosine schedule. Val loss 2.31 → 1.87. See
+`research/distill_students/train_marketing.py` in the app repo.
+
+## License
+
+Pleius internal — see https://pleius.com. Not for redistribution.

config.json

@@ -0,0 +1,20 @@
+{
+  "modality": "marketing",
+  "encoder": "openai/clip-vit-base-patch32",
+  "encoder_dim": 768,
+  "n_pillars": 4,
+  "pillars": [
+    "universal_appeal",
+    "demographic_appeal",
+    "audience_drive",
+    "engagement"
+  ],
+  "d_model": 256,
+  "n_layers": 4,
+  "max_seq_len": 16,
+  "vocab_size": 115,
+  "bos_id": 1,
+  "eos_id": 2,
+  "pad_id": 0,
+  "sep_id": 3
+}

student_int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2cb29dbdbdd5431d927b52a0d92fb6208085958e787cc14d8755a50c1eaed04
+size 7226461

tokenizer.json

@@ -0,0 +1,120 @@
+{
+  "modality": "marketing",
+  "tokens": [
+    "<pad>",
+    "<bos>",
+    "<eos>",
+    "<sep>",
+    "eye catching",
+    "scroll stopping",
+    "thumb stopping",
+    "pops on feed",
+    "stops the scroll",
+    "bold colors",
+    "good contrast",
+    "clean composition",
+    "strong focal point",
+    "good lighting",
+    "well lit",
+    "well framed",
+    "color works",
+    "color palette slaps",
+    "vibe is right",
+    "looks premium",
+    "looks expensive",
+    "feels intentional",
+    "too busy",
+    "blurry",
+    "low contrast",
+    "no clear focus",
+    "bad lighting",
+    "cluttered",
+    "looks dated",
+    "looks like 2014",
+    "uncanny",
+    "ai generated feel",
+    "weird crop",
+    "off center weird",
+    "background too loud",
+    "colors clash",
+    "low effort",
+    "lazy design",
+    "no vibe",
+    "looks like clip art",
+    "on brand",
+    "feels native to the platform",
+    "looks like a real photo",
+    "model looks natural",
+    "feels like a real creator",
+    "talks to the right person",
+    "knows the audience",
+    "feels organic",
+    "doesn't feel like an ad",
+    "lands for the target",
+    "the right energy",
+    "right vibe for the audience",
+    "off brand",
+    "screams ad",
+    "looks like an ad",
+    "stock photo feel",
+    "feels like a stock photo",
+    "wrong audience",
+    "wrong tone",
+    "feels generic",
+    "feels templated",
+    "model looks fake",
+    "wrong vibe",
+    "doesn't fit the platform",
+    "clear cta",
+    "the offer pops",
+    "price tag clear",
+    "deal feels real",
+    "social proof shows",
+    "you'd actually click",
+    "you know what they sell",
+    "product is the hero",
+    "hero shot works",
+    "instantly readable",
+    "headline lands",
+    "headline sells it",
+    "weak cta",
+    "no offer",
+    "offer is unclear",
+    "can't read the cta",
+    "where's the product",
+    "what are they selling",
+    "headline buried",
+    "headline doesn't sell",
+    "no reason to click",
+    "small text",
+    "small cta",
+    "the ask is buried",
+    "clear product shot",
+    "clear text",
+    "memorable",
+    "saveable",
+    "shareable",
+    "you'd save this",
+    "you'd send this to a friend",
+    "feels native",
+    "screams brand",
+    "logo placement good",
+    "logo readable",
+    "text hierarchy clean",
+    "tight crop",
+    "negative space works",
+    "too much text",
+    "wall of text",
+    "cluttered text",
+    "no hierarchy",
+    "logo is huge",
+    "logo is invisible",
+    "you'd scroll past",
+    "forgettable",
+    "boring",
+    "low contrast text",
+    "text overlaps the product",
+    "background fights the product",
+    "looks clean"
+  ]
+}