Add model card

MODEL_CARD.md

@@ -1,4 +1,5 @@
 ---
+---
 language: English
 license: apache-2.0
 tags:
@@ -6,23 +7,42 @@ tags:
   - bert
   - fine-tuned
   - recipe-bot
-datasets: []
-metrics: []
+  - demo
+datasets:
+  - local: data/train.csv
+metrics:
+  - name: train_loss
+    type: scalar
+    value: 0.622  # example from 1-epoch run
+model-index:
+  - name: recipe-bert
+    results:
+      - task: text-classification
+        dataset: local/data/train.csv
+        metrics:
+          - name: train_loss
+            type: scalar
+            value: 0.622
 ---
 
 # recipe-bert
 
-A small fine-tuned BERT model for demonstration (trained on a tiny sample). Replace with your real dataset for production.
+Short description
+A small demo fine-tuned BERT model intended for learning and prototyping. Trained on a tiny synthetic dataset; not suitable for production.
 
-## Model description
+Training details
 - Base model: bert-base-uncased
 - Task: binary text classification
-- Training: quick 1-epoch smoke run (demo)
+- Dataset: local CSV `data/train.csv` (columns: `text,label`)
+- Hyperparameters (demo):
+  - epochs: 1
+  - batch size: 8
+  - optimizer: AdamW (default)
+  - seed: 42
 
-## How to use
+How to use
 ```python
 from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline
-
 model_id = "x2-world/recipe-bert"
 tokenizer = AutoTokenizer.from_pretrained(model_id)
 model = AutoModelForSequenceClassification.from_pretrained(model_id)
@@ -30,5 +50,6 @@ clf = pipeline('text-classification', model=model, tokenizer=tokenizer)
 print(clf('The pizza was great'))
 ```
 
-## Limitations
-This model was trained on an extremely small synthetic dataset—do not use in production.
+Limitations and license
+- Trained on a toy dataset. Expect poor performance on real data.
+- License: Apache-2.0