Upload 3 files

Apologies for the late upload of the run_inference.py and other necessary files for the functional inference and model verification.
5 :00 AM inspiration woke me up.

emotion_model.py

@@ -0,0 +1,52 @@
+# emotion_model.py
+# ROSA: Recursive Ontology of Semantic Affect
+# Sublime Emotional System by Willinton Triana Cardona
+
+from transformers import BertModel
+import torch.nn as nn
+import torch
+
+class Rosa(nn.Module):
+    def __init__(
+        self,
+        model_name="bert-base-uncased",
+        num_emotions=29,
+        latent_dim=None,
+        return_vector=False,
+        emotion_labels=None
+    ):
+        super().__init__()
+        self.heart = BertModel.from_pretrained(model_name)
+        self.grace = nn.Dropout(0.3)
+        self.bloom = nn.Linear(self.heart.config.hidden_size, num_emotions)
+
+        self.return_vector = return_vector
+        self.latent_dim = latent_dim
+        self.emotion_labels = emotion_labels or [
+            "admiration", "amusement", "anger", "annoyance", "approval", "caring", "confusion", "curiosity",
+            "desire", "disappointment", "disapproval", "disgust", "embarrassment", "excitement", "fear",
+            "gratitude", "grief", "joy", "love", "nervousness", "optimism", "pride", "realization", "relief",
+            "remorse", "sadness", "surprise", "neutral"
+        ]
+
+        if latent_dim:
+            self.rosa_embedding = nn.Linear(self.heart.config.hidden_size, latent_dim)
+
+        self.loss_fct = nn.BCEWithLogitsLoss()
+
+    def forward(self, input_ids, attention_mask, labels=None):
+        petals = self.heart(input_ids=input_ids, attention_mask=attention_mask)
+        pooled = petals.pooler_output
+        softened = self.grace(pooled)
+
+        if self.return_vector and hasattr(self, 'rosa_embedding'):
+            embedding = self.rosa_embedding(softened)
+            return {"embedding": embedding}
+
+        logits = self.bloom(softened)
+
+        if labels is not None:
+            loss = self.loss_fct(logits, labels.float())
+            return {"loss": loss, "logits": logits}
+        else:
+            return {"logits": logits}

requirements.txt

@@ -0,0 +1,3 @@
+transformers>=4.30.0
+torch>=2.0.0
+scikit-learn

run_inference.py

@@ -0,0 +1,29 @@
+# run_inference.py
+# Simple test script to evaluate ROSA on custom input
+
+from transformers import BertTokenizer
+from emotion_model import Rosa
+import torch
+
+def predict(text):
+    tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
+    emotion_labels = [
+        "admiration", "amusement", "anger", "annoyance", "approval", "caring", "confusion", "curiosity",
+        "desire", "disappointment", "disapproval", "disgust", "embarrassment", "excitement", "fear",
+        "gratitude", "grief", "joy", "love", "nervousness", "optimism", "pride", "realization", "relief",
+        "remorse", "sadness", "surprise", "neutral"
+    ]
+
+    model = Rosa(num_emotions=29, emotion_labels=emotion_labels)
+    model.load_state_dict(torch.load("rosa.pt", map_location=torch.device("cpu")))
+    model.eval()
+
+    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
+    with torch.no_grad():
+        logits = model(**inputs)['logits']
+        probs = torch.sigmoid(logits).squeeze()
+
+    return {label: float(prob) for label, prob in zip(emotion_labels, probs)}
+
+if __name__ == "__main__":
+    print(predict("My heart is filled with longing and beauty."))