Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -249,8 +249,14 @@ class XylariaChat:
|
|
249 |
|
250 |
def query_knowledge_graph(self, query):
|
251 |
query_embedding = self.embedding_model.encode(query, convert_to_tensor=True)
|
252 |
-
|
253 |
-
node_embeddings = {
|
|
|
|
|
|
|
|
|
|
|
|
|
254 |
|
255 |
similarities = {node: util.pytorch_cos_sim(query_embedding, embedding)[0][0].item() for node, embedding in node_embeddings.items()}
|
256 |
|
@@ -340,27 +346,17 @@ class XylariaChat:
|
|
340 |
return f"Error during Math OCR: {e}"
|
341 |
|
342 |
def extract_entities_and_relations(self, text):
|
343 |
-
|
344 |
-
|
345 |
-
with torch.no_grad():
|
346 |
-
outputs = self.embedding_model(**doc)
|
347 |
-
|
348 |
entities = []
|
349 |
relations = []
|
350 |
-
for i in range(len(doc['input_ids'][0])):
|
351 |
-
token = self.embedding_model.tokenizer.decode(doc['input_ids'][0][i])
|
352 |
-
if outputs['last_hidden_state'][0][i].norm() > 3:
|
353 |
-
entities.append(token)
|
354 |
|
355 |
-
|
356 |
-
|
357 |
-
relation = f"{entities[i]} related_to {entities[i+1]}"
|
358 |
-
relations.append(relation)
|
359 |
|
360 |
return entities, relations
|
361 |
|
362 |
-
def update_knowledge_graph(self,
|
363 |
-
entities, relations = self.extract_entities_and_relations(text)
|
364 |
for entity in entities:
|
365 |
self.knowledge_graph.add_node(entity)
|
366 |
for relation in relations:
|
@@ -372,7 +368,8 @@ class XylariaChat:
|
|
372 |
|
373 |
def get_response(self, user_input, image=None):
|
374 |
try:
|
375 |
-
self.
|
|
|
376 |
|
377 |
messages = []
|
378 |
|
|
|
249 |
|
250 |
def query_knowledge_graph(self, query):
|
251 |
query_embedding = self.embedding_model.encode(query, convert_to_tensor=True)
|
252 |
+
|
253 |
+
node_embeddings = {}
|
254 |
+
for node in self.knowledge_graph.nodes():
|
255 |
+
try:
|
256 |
+
node_embedding = self.embedding_model.encode(node, convert_to_tensor=True)
|
257 |
+
node_embeddings[node] = node_embedding
|
258 |
+
except Exception as e:
|
259 |
+
print(f"Error encoding node {node}: {e}")
|
260 |
|
261 |
similarities = {node: util.pytorch_cos_sim(query_embedding, embedding)[0][0].item() for node, embedding in node_embeddings.items()}
|
262 |
|
|
|
346 |
return f"Error during Math OCR: {e}"
|
347 |
|
348 |
def extract_entities_and_relations(self, text):
|
349 |
+
inputs = self.embedding_model.tokenizer(text, return_tensors="pt", padding=True, truncation=True)
|
350 |
+
|
|
|
|
|
|
|
351 |
entities = []
|
352 |
relations = []
|
|
|
|
|
|
|
|
|
353 |
|
354 |
+
entities, relations = self.extract_entities_and_relations(message)
|
355 |
+
self.update_knowledge_graph(entities, relations)
|
|
|
|
|
356 |
|
357 |
return entities, relations
|
358 |
|
359 |
+
def update_knowledge_graph(self, entities, relations):
|
|
|
360 |
for entity in entities:
|
361 |
self.knowledge_graph.add_node(entity)
|
362 |
for relation in relations:
|
|
|
368 |
|
369 |
def get_response(self, user_input, image=None):
|
370 |
try:
|
371 |
+
entities, relations = self.extract_entities_and_relations(user_input)
|
372 |
+
self.update_knowledge_graph(entities, relations)
|
373 |
|
374 |
messages = []
|
375 |
|