Add ORD forward reaction prediction demo app

.gitignore

@@ -0,0 +1,26 @@
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+.env
+.venv
+env/
+venv/
+ENV/
+.DS_Store

README.md

@@ -1,13 +1,58 @@
 ---
-title: Demo Space Forward
-emoji: 📉
-colorFrom: gray
-colorTo: blue
+title: ORD Forward Reaction Prediction
+emoji: 🧪
+colorFrom: blue
+colorTo: green
 sdk: gradio
 sdk_version: 5.49.1
 app_file: app.py
 pinned: false
-license: mit
+license: apache-2.0
+models:
+  - smitathkr1/ord-forward-t5
+datasets:
+  - smitathkr1/ord-reactions
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# ORD Forward Reaction Prediction - T5 Model
+
+This is a demo space for testing the `smitathkr1/ord-forward-t5` model, which predicts chemical reaction products from reactants.
+
+## Model Details
+
+- **Model**: T5-based forward reaction prediction model
+- **Training Data**: 2.3M reactions from the Open Reaction Database (ORD)
+- **Training Status**: 5 epochs completed
+- **Dataset**: [smitathkr1/ord-reactions](https://huggingface.co/datasets/smitathkr1/ord-reactions)
+
+## Usage
+
+Enter reactants in SMILES format (e.g., `CC(C)N1CCN(C)CC1.Brc1ccccc1`) and the model will predict the product.
+
+### Input Format
+- Reactants should be in SMILES notation
+- Multiple reactants can be separated by '.'
+- Example: `amine.aryl_halide` → `product`
+
+### Parameters
+- **Max Length**: Maximum length of the generated product SMILES
+- **Num Beams**: Number of beams for beam search (higher = more thorough search)
+- **Temperature**: Sampling temperature (higher = more diverse outputs)
+
+## Examples
+
+The model was trained on various reaction types from the ORD database, including:
+- Buchwald-Hartwig amination reactions
+- Cross-coupling reactions
+- And many other organic reactions
+
+## Citation
+
+If you use this model, please cite:
+- The Open Reaction Database: https://open-reaction-database.org/
+- Model: https://huggingface.co/smitathkr1/ord-forward-t5
+- Dataset: https://huggingface.co/datasets/smitathkr1/ord-reactions
+
+## License
+
+Apache 2.0

app.py

@@ -0,0 +1,143 @@
+import gradio as gr
+from transformers import T5ForConditionalGeneration, T5Tokenizer
+import torch
+
+# Load the model and tokenizer
+model_name = "smitathkr1/ord-forward-t5"
+print(f"Loading model: {model_name}")
+tokenizer = T5Tokenizer.from_pretrained(model_name)
+model = T5ForConditionalGeneration.from_pretrained(model_name)
+
+# Set device
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+model.to(device)
+model.eval()
+
+def predict_reaction(reactants_smiles, max_length=150, num_beams=5, temperature=1.0):
+    """
+    Predict the product of a chemical reaction given reactants in SMILES format.
+    
+    Args:
+        reactants_smiles: SMILES string of reactants (can be separated by '.')
+        max_length: Maximum length of generated sequence
+        num_beams: Number of beams for beam search
+        temperature: Sampling temperature
+    
+    Returns:
+        Predicted product SMILES
+    """
+    try:
+        # Prepare input
+        input_text = reactants_smiles.strip()
+        
+        # Tokenize input
+        inputs = tokenizer(
+            input_text,
+            return_tensors="pt",
+            max_length=512,
+            truncation=True,
+            padding=True
+        ).to(device)
+        
+        # Generate prediction
+        with torch.no_grad():
+            outputs = model.generate(
+                inputs["input_ids"],
+                max_length=max_length,
+                num_beams=num_beams,
+                temperature=temperature,
+                early_stopping=True,
+                do_sample=False
+            )
+        
+        # Decode output
+        predicted_product = tokenizer.decode(outputs[0], skip_special_tokens=True)
+        
+        return predicted_product
+    
+    except Exception as e:
+        return f"Error: {str(e)}"
+
+# Example inputs from ORD dataset
+examples = [
+    ["CC(C)N1CCN(C)CC1.Brc1ccccc1"],  # Buchwald-Hartwig amination example
+    ["CCN1CCNCC1.Ic1ccccc1"],  # Another coupling reaction
+    ["CC(=O)N1CCNCC1.Clc1ccccc1"],  # Chloro coupling
+]
+
+# Create Gradio interface
+iface = gr.Interface(
+    fn=predict_reaction,
+    inputs=[
+        gr.Textbox(
+            label="Reactants (SMILES)",
+            placeholder="Enter reactants in SMILES format (e.g., CC(C)N1CCN(C)CC1.Brc1ccccc1)",
+            lines=2
+        ),
+        gr.Slider(
+            minimum=50,
+            maximum=300,
+            value=150,
+            step=10,
+            label="Max Length"
+        ),
+        gr.Slider(
+            minimum=1,
+            maximum=10,
+            value=5,
+            step=1,
+            label="Num Beams"
+        ),
+        gr.Slider(
+            minimum=0.1,
+            maximum=2.0,
+            value=1.0,
+            step=0.1,
+            label="Temperature"
+        )
+    ],
+    outputs=gr.Textbox(
+        label="Predicted Product (SMILES)",
+        lines=2
+    ),
+    examples=examples,
+    title="🧪 ORD Forward Reaction Prediction - T5 Model",
+    description="""
+    ## Forward Reaction Prediction using T5
+    
+    This model predicts chemical reaction products from reactants using a T5 model trained on 2.3M reactions from the Open Reaction Database (ORD).
+    
+    **Model:** `smitathkr1/ord-forward-t5` (5 epochs completed)
+    
+    **Dataset:** [smitathkr1/ord-reactions](https://huggingface.co/datasets/smitathkr1/ord-reactions)
+    
+    ### How to use:
+    1. Enter reactants in SMILES format (separate multiple reactants with '.')
+    2. Adjust generation parameters if needed
+    3. Click Submit to get the predicted product
+    
+    ### Example reactions:
+    - Buchwald-Hartwig amination reactions
+    - Various coupling reactions from the ORD database
+    """,
+    article="""
+    ### About the Model
+    This T5 model was trained on 2.3 million reactions from the Open Reaction Database.
+    The training has completed 5 epochs so far.
+    
+    ### Citation
+    If you use this model, please cite the Open Reaction Database:
+    - [Open Reaction Database](https://open-reaction-database.org/)
+    
+    ### Notes
+    - Input should be valid SMILES strings
+    - The model predicts forward reactions (reactants → products)
+    - Adjust beam search parameters for different prediction strategies
+    """,
+    theme=gr.themes.Soft(),
+    allow_flagging="never"
+)
+
+# Launch the app
+if __name__ == "__main__":
+    iface.launch()

requirements.txt

@@ -0,0 +1,6 @@
+transformers>=4.30.0
+torch>=2.0.0
+gradio>=4.0.0
+sentencepiece
+protobuf
+accelerate