upload cortexa-create-feedback v1

README.md

@@ -0,0 +1,63 @@
+---
+language:
+- en
+license: other
+license_name: pleius-internal
+tags:
+- onnx
+- conditional-text-generation
+- video-feedback
+- distillation
+- creator-tools
+---
+
+# cortexa-create-feedback (distilled student)
+
+A ~4.4M-parameter conditional decoder distilled from
+`M725/cortexa-create-scorer` outputs. Takes CLIP-ViT-B/32 vision
+features (mean-pooled across video frames, 768-d) + the 5 Create pillar
+scores and emits a creator-vernacular phrase chain about the short-form
+video:
+
+```
+"first frame slaps | feels intentional"
+"thumb stopping | shareable"
+"filler | feels rushed | first frame is nothing"
+"feels off beat | slow open | no payoff"
+```
+
+## 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 ~138; specials `<pad>`, `<bos>`, `<eos>`, `<sep>`). |
+| `config.json`       | Encoder dim, pillar names, vocab size, special-token ids. |
+
+## Inference shape
+
+```
+inputs:
+  encoder_feats   (1, 768)  float32   # mean-pooled CLIP-ViT-B/32 vision across frames
+  scores          (1, 5)    float32   # [hook, hold, algorithmic_fit, brand_lift, overall] in [0,1]
+  scores_present  (1,)      float32   # 1.0 anchored, 0.0 fast-mode
+  input_ids       (1, T)    int64
+outputs:
+  logits          (1, T, V) float32
+```
+
+Same sampling recommendation as `cortexa-marketing-feedback`: temperature
+0.8 + top-k 20 + SEP-veto.
+
+## Training
+
+6k phrase triples from 3 real short-form videos
+(`public/create-tutorial/*.mp4`) + 1997 synthetic "videos" built by
+random-crop + color jitter over COCO stills (each frame goes through
+cortexa_v10 separately, so the per-frame curve has real variation). 15
+epochs. Val loss 2.39 → 1.97. See
+`research/distill_students/train_create.py` in the app repo.
+
+## License
+
+Pleius internal — see https://pleius.com. Not for redistribution.

config.json

@@ -0,0 +1,21 @@
+{
+  "modality": "create",
+  "encoder": "openai/clip-vit-base-patch32",
+  "encoder_dim": 768,
+  "n_pillars": 5,
+  "pillars": [
+    "hook",
+    "hold",
+    "algorithmic_fit",
+    "brand_lift",
+    "overall"
+  ],
+  "d_model": 256,
+  "n_layers": 4,
+  "max_seq_len": 16,
+  "vocab_size": 138,
+  "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 sha256:2fd3f324a4cf5d14ca7db136981b49070672411c8c5dc479e4d23922b91ff67b
+size 7238496

tokenizer.json

@@ -0,0 +1,143 @@
+{
+  "modality": "create",
+  "tokens": [
+    "<pad>",
+    "<bos>",
+    "<eos>",
+    "<sep>",
+    "fast paced hook",
+    "hook lands",
+    "opening lands",
+    "first frame slaps",
+    "first 2 seconds work",
+    "good first impression",
+    "actually grabs you",
+    "i kept watching",
+    "scroll stopping",
+    "thumb stopping",
+    "strong opening",
+    "first 3 seconds work",
+    "hook didn't land",
+    "first 3 seconds dead",
+    "slow open",
+    "slow start",
+    "buried the hook",
+    "i'd skip",
+    "i'd swipe",
+    "first frame is nothing",
+    "weak opening",
+    "no reason to watch",
+    "where's the hook",
+    "i'm gone in 1 second",
+    "engaging movement",
+    "tight editing",
+    "great pacing",
+    "no dead air",
+    "every cut earns it",
+    "kept me to the end",
+    "stayed locked in",
+    "high energy",
+    "kept the energy",
+    "good rhythm",
+    "didn't skip",
+    "didn't swipe",
+    "lost me in the middle",
+    "drags",
+    "drags in the middle",
+    "filler shots",
+    "filler",
+    "too long",
+    "could be 10 seconds",
+    "boring middle",
+    "low energy",
+    "no rhythm",
+    "feels long",
+    "i bounced halfway",
+    "smooth transitions",
+    "good music sync",
+    "music matches",
+    "music elevates",
+    "beat drop works",
+    "good lighting",
+    "lighting clean",
+    "color grade clean",
+    "captions on point",
+    "captions readable",
+    "feels native",
+    "feels organic",
+    "looks like a real creator",
+    "looks high production",
+    "looks pro",
+    "framing works",
+    "shots feel intentional",
+    "too many cuts",
+    "shaky camera",
+    "shaky",
+    "out of focus",
+    "looks blurry",
+    "looks pixelated",
+    "out of sync",
+    "off beat",
+    "cuts feel random",
+    "music feels random",
+    "music too loud",
+    "music kills the vo",
+    "captions wrong timing",
+    "captions cover face",
+    "vertical crop weird",
+    "wrong aspect",
+    "lighting bad",
+    "shadows weird",
+    "looks dark",
+    "looks overexposed",
+    "clear voiceover",
+    "audio is clean",
+    "audio is crisp",
+    "mic is good",
+    "voiceover clear",
+    "satisfying ending",
+    "payoff lands",
+    "ending pays off",
+    "loops well",
+    "good loop",
+    "would save this",
+    "would share this",
+    "would rewatch",
+    "doesn't feel like an ad",
+    "muddy audio",
+    "background noise",
+    "audio peaks",
+    "audio clipping",
+    "mic is bad",
+    "echo",
+    "wind noise",
+    "abrupt ending",
+    "weak ending",
+    "ending falls flat",
+    "no payoff",
+    "doesn't loop",
+    "ai voice",
+    "ai face",
+    "uncanny",
+    "screams ad",
+    "looks like a stock video",
+    "low effort",
+    "feels intentional",
+    "every shot earns it",
+    "the vibe is right",
+    "saveable",
+    "shareable",
+    "would send this to a friend",
+    "forgettable",
+    "boring",
+    "no vibe",
+    "feels random",
+    "feels rushed",
+    "feels stitched together",
+    "would scroll past",
+    "fast hook",
+    "loses you halfway",
+    "feels off beat",
+    "no hook"
+  ]
+}