Add automatic model card generation with dataset documentation

app.py

@@ -327,6 +327,111 @@ def train_model(
         yield from log("✅ TRAINING COMPLETED!")
         yield from log("=" * 80)
         yield from log(f"📊 Final Loss: {result.training_loss:.4f}")
+        
+        # Generate model card with dataset info
+        yield from log("")
+        yield from log("📝 Generating model card...")
+        
+        from datetime import datetime
+        
+        # Build dataset info for model card
+        dataset_info = []
+        dataset_hf_ids = []
+        for dataset_name in selected_datasets:
+            if " / " in dataset_name:
+                dataset_short_name = dataset_name.split(" / ", 1)[1]
+            else:
+                dataset_short_name = dataset_name
+            
+            for category in DATASETS.values():
+                if dataset_short_name in category:
+                    ds_config = category[dataset_short_name]
+                    dataset_info.append(f"- [{dataset_short_name}](https://huggingface.co/datasets/{ds_config['hf_id']}) ({ds_config['size']})")
+                    dataset_hf_ids.append(ds_config['hf_id'])
+                    break
+        
+        model_card = f"""---
+language:
+- en
+license: apache-2.0
+tags:
+- zen
+- vision-language
+- function-calling
+- agent
+base_model: {model_config['hf_id']}
+datasets:
+{chr(10).join([f"- {hf_id}" for hf_id in dataset_hf_ids])}
+---
+
+# {output_repo.split('/')[-1]}
+
+Fine-tuned from [{model_config['hf_id']}](https://huggingface.co/{model_config['hf_id']}) using the Zen Training Space.
+
+## Training Details
+
+### Base Model
+- **Model**: {model_short_name}
+- **Size**: {model_config['size']} parameters
+- **Type**: {model_config['type']}
+- **Base HF ID**: [{model_config['hf_id']}](https://huggingface.co/{model_config['hf_id']})
+
+### Datasets Used
+{chr(10).join(dataset_info)}
+
+### Training Configuration
+- **Total Samples**: {len(all_datasets):,}
+- **Epochs**: {epochs}
+- **Batch Size**: {batch_size}
+- **Learning Rate**: {learning_rate}
+- **Precision**: bfloat16
+- **Final Training Loss**: {result.training_loss:.4f}
+- **Training Date**: {datetime.now().strftime('%Y-%m-%d %H:%M:%S UTC')}
+
+### Hardware
+- **GPU**: NVIDIA A10G (24GB)
+- **Platform**: HuggingFace Spaces
+
+## Usage
+
+```python
+from transformers import AutoModel, AutoProcessor
+
+model = AutoModel.from_pretrained("{output_repo}", trust_remote_code=True)
+processor = AutoProcessor.from_pretrained("{output_repo}")
+
+# Your inference code here
+```
+
+## Training Space
+
+This model was trained using the [Zen Training Space](https://huggingface.co/spaces/zeekay/zen-training), 
+a unified platform for training all Zen AI models.
+
+## Citation
+
+```bibtex
+@misc{{{output_repo.replace('/', '_').replace('-', '_')},
+  author = {{Zen AI}},
+  title = {{{output_repo.split('/')[-1]}}},
+  year = {{2025}},
+  publisher = {{HuggingFace}},
+  url = {{https://huggingface.co/{output_repo}}}
+}}
+```
+
+---
+
+*Trained with ❤️ using [Zen Training Space](https://huggingface.co/spaces/zeekay/zen-training)*
+"""
+        
+        # Save model card
+        import os
+        os.makedirs("./training-output", exist_ok=True)
+        with open("./training-output/README.md", "w") as f:
+            f.write(model_card)
+        
+        yield from log("✅ Model card generated")
         yield from log(f"☁️  Model uploaded to: {output_repo}")
         yield from log("")
         yield from log("🎉 SUCCESS!")