card: training described at outcome level only

README.md

@@ -85,9 +85,9 @@ Plain bi-encoder: one 768-d vector per image, one per query, cosine similarity.
 
 - **Base:** SigLIP2-base-384 (`timm/ViT-B-16-SigLIP2-384`), full fine-tune of both towers.
 - **Size — 213,159,938 params:** the base's 256K-token multilingual vocabulary is trimmed to the 44,646 pieces fashion text actually uses (BPE merge-closure preserved). Verified: **100.00% bit-identical** outputs on 12,000 benchmark queries and 6,000 held-out registers vs the untrimmed 375M model — every benchmark number in this card transfers exactly. Out-of-vocabulary input (emoji, non-Latin scripts, rare typos) re-segments gracefully into kept pieces (cosine 0.82–0.97 to full model), never a silent failure.
-- **Data:** 93K commercial product images (from [Marqo-GS-10M](https://huggingface.co/datasets/Marqo/marqo-GS-10M), Apache-2.0) re-annotated by a frontier VLM into structured attributes and four query registers, with graded relevance labels (99.9% coverage).
-- **Recipe:** graded contrastive learning over multiple query registers, an anchor loss to the frozen base to prevent catastrophic transfer loss, and weight-space interpolation for the released checkpoint. The training pipeline and dataset are not released.
-- **Distinctive property:** unlike typical fashion fine-tunes, Pro Lite does **not** collapse out of domain — the anchored recipe held generalist behavior while adding fashion skill (image-tower drift from base: 0.36%).
+- **Data:** 93K commercial product images (from [Marqo-GS-10M](https://huggingface.co/datasets/Marqo/marqo-GS-10M), Apache-2.0), re-annotated by a frontier VLM with graded relevance labels (99.9% coverage).
+- **Training:** a proprietary fine-tuning recipe; the pipeline and annotated dataset are not released.
+- **Distinctive property:** unlike typical fashion fine-tunes, Pro Lite does **not** collapse out of domain — it adds fashion skill while keeping generalist behavior (image-tower drift from base: 0.36%).
 
 ## Evaluation protocol
 
@@ -97,7 +97,7 @@ Measured on six academic fashion benchmarks (DeepFashion In-Shop, DeepFashion Mu
 
 ```bibtex
 @misc{modaprolite2026,
-  title  = {MODA Pro Lite: an anchored fashion encoder that does not collapse out of domain},
+  title  = {MODA Pro Lite: a fashion encoder that does not collapse out of domain},
   author = {Hopit AI},
   year   = {2026},
   url    = {https://huggingface.co/HopitAI/moda-pro-lite}