Replace starter README with full model card

README.md

@@ -1,51 +1,39 @@
 ---
-license: other
-library_name: transformers
+license: apache-2.0
+base_model: google/embeddinggemma-300m
+library_name: sklearn
 tags:
 - conference-demo
 - local-ai
-- fine-tuning
-- gemma
-- qwen
-- thinktecture
+- intent-classification
+- logistic-regression
 ---
 
-# LogReg intent classifier (on top of EmbeddingGemma — Nextera demo)
-
 > ⚠️ **Conference talk demo — not production weights.**
 >
-> This model accompanies a conference keynote on local on-device AI. It is
-> published as a reference for the fine-tuning patterns shown on stage,
-> **not** as a deployable artefact. No security audit, no SLA, pinned to
-> the talk's state.
+> This model accompanies a conference keynote on local on-device AI. Published
+> as a reference for the fine-tuning patterns shown on stage — **not** a
+> deployable artefact. No security audit, no SLA, pinned to the talk's state.
 >
-> Source repository:
-> [thinktecture-labs/local-multi-model-agent-slm](https://github.com/thinktecture-labs/local-multi-model-agent-slm)
-> Threat model + out-of-scope items:
-> [`SECURITY.md`](https://github.com/thinktecture-labs/local-multi-model-agent-slm/blob/main/SECURITY.md)
+> - Source repository: [thinktecture-labs/local-multi-model-agent-slm](https://github.com/thinktecture-labs/local-multi-model-agent-slm)
+> - Threat model + out-of-scope: [`SECURITY.md`](https://github.com/thinktecture-labs/local-multi-model-agent-slm/blob/main/SECURITY.md)
+> - All five models in the stack: [Collection — Local Multi-Model Agent — nextera fine-tunes](https://huggingface.co/collections/thinktecture/local-multi-model-agent-nextera-fine-tunes-6a04a8ff2a40e5696f3c2f18)
 
 ---
 
-## What this is
-
-Fine-tune of [`google/embeddinggemma-300m (via fine-tuned embeddings)`](https://huggingface.co/thinktecture/embeddinggemma-300m-ft-nextera) for the demo's reference scenario
-("Nextera" — a fully synthetic SaaS analytics product invented for the talk).
-
-See [`finetune/MODEL_CARDS.md#LogReg`](https://github.com/thinktecture-labs/local-multi-model-agent-slm/blob/main/finetune/MODEL_CARDS.md#logreg)
-in the source repository for the full card — training data, hyperparameters,
-eval scores, known failure modes.
-
-## License
+## LogReg Intent Classifier
+
+| | |
+|---|---|
+| **Base** | scikit-learn `LogisticRegression`, multinomial, L2 penalty |
+| **License** | Apache-2.0 (this repo) — but inputs are EmbeddingGemma vectors so the [Gemma Terms](https://github.com/thinktecture-labs/local-multi-model-agent-slm/blob/main/finetune/MODEL_LICENSES.md) cover the embedding step |
+| **Training script** | [`training/train_intent_logreg.py`](https://github.com/thinktecture-labs/local-multi-model-agent-slm/blob/main/training/train_intent_logreg.py) |
+| **Method** | LogReg on FT-EmbeddingGemma's 768-dim output vectors. Held-out 90/10 split. ~2 minutes on CPU. |
+| **Training data** | Same as Gemma3-1B intent: `data/training-data/gemma3_intent_{scenario}.jsonl` (re-embedded with the FT EmbeddingGemma) |
+| **Hardware** | CPU is sufficient. Requires the FT EmbeddingGemma llama-server running on port 9092/9096 to embed training examples. |
+| **Intended use** | Replaces the 1B generative classifier as the primary intent router. ~10ms per query (vs ~200ms for the 1B). Same accuracy on the standard eval set. |
+| **Out of scope** | Anything that requires generation (it's a 3-way classifier). Falls back to the 1B classifier when confidence is below threshold (defaults to 0.65). |
+| **Reference eval (Nextera)** | 96.1% on 180-query eval set. ~10ms per classification (single CPU thread). |
+| **Known failure modes** | When the EmbeddingGemma FT changes, the LogReg weights become invalid — `intent_classifier_logreg.py:13-15` warns about this coupling. Re-train both together. |
 
-This artefact is a derivative of [`google/embeddinggemma-300m (via fine-tuned embeddings)`](https://huggingface.co/thinktecture/embeddinggemma-300m-ft-nextera) and inherits
-its license: **Apache-2.0 (this artifact) + Gemma Terms (for the embedding step)**. See
-[`finetune/MODEL_LICENSES.md`](https://github.com/thinktecture-labs/local-multi-model-agent-slm/blob/main/finetune/MODEL_LICENSES.md)
-for the full per-model license summary.
-
-## Collection
-
-This model is part of the
-[Local Multi-Model Agent — nextera fine-tunes](https://huggingface.co/collections/thinktecture/local-multi-model-agent-nextera-fine-tunes-6a04a8ff2a40e5696f3c2f18)
-collection — five models in the production stack: intent (Gemma 1B), retrieval
-(EmbeddingGemma), tool calling (Qwen 4B), RAG synthesis (Gemma 4B), and the
-LogReg intent classifier.
+---