update

app.py

@@ -28,7 +28,7 @@ import spaces
 #property_names = list(candidate_models.keys())
 model = MolecularGenerationModel()
 
-@spaces.GPU(duration=60)
+@spaces.GPU(duration=120)
 def predict_single_label(logp, tpas, sas, qed, logp_choose, tpsa_choose, sas_choose, qed_choose):
     input_dict = dict()
     if logp_choose:

llama_customized_models.py

@@ -130,9 +130,7 @@ class LlamaForCausalLMWithNumericalEmbedding(LlamaForCausalLM):
         assert len(properties) == b, "The number of properties should be equal to the batch size."
         assert len(properties_index) == b, "The number of properties_index should be equal to the batch size."
 
-        print(input_ids, "input_ids")
         embeddings = self.model.embed_tokens(input_ids)
-        print(embeddings, "embeddings")
         
         for i, (props, props_index, embeds) in enumerate(zip(properties, properties_index, embeddings)):
             assert len(props) == len(props_index), "The number of properties should be equal to the number of properties_index."

utils.py

@@ -287,9 +287,7 @@ class MolecularGenerationModel():
             with torch.set_grad_enabled(False):
                 early_stop_flags = torch.zeros(num_generations, dtype=torch.bool).to(self.model.device)
                 for k in range(steps):
-                    print("batch", batch)
                     logits = self.model(**batch)['logits']
-                    print("logits", logits)
                     logits = logits[:, -1, :] / temperature
                     probs = F.softmax(logits, dim=-1)
                     ix = torch.multinomial(probs, num_samples=num_generations)
@@ -354,8 +352,8 @@ class MolecularGenerationModel():
         # delete the condition columns
         new_df = new_df.drop(columns=[col for col in new_df.columns if "condition" in col])
 
-        # drop the empty smiles rows
-        new_df = new_df.dropna(subset=['SMILES'])
+        # drop the rows 
+        df = df[df["SMILES"] != ""]
 
         # convert the measured to 2 decimal places
         new_df = new_df.round(2)