Update to make it compatible with new OpenAI SDK

src/ai.py

@@ -33,6 +33,13 @@ def embeddings(texts, **kw):
 	model = kw.get('model','text-embedding-ada-002')
 	llm = openai.model(model)
 	resp = llm.embed_many(texts, **kw)
+	# Convert usage object to dictionary if needed
+	if hasattr(resp['usage'], 'prompt_tokens'):
+		resp['usage'] = {
+			'prompt_tokens': resp['usage'].prompt_tokens,
+			'completion_tokens': resp['usage'].completion_tokens,
+			'total_tokens': resp['usage'].total_tokens
+		}
 	resp['model'] = model
 	return resp
 
@@ -42,7 +49,17 @@ def get_token_count(text):
 
 def stats_callback(out, resp, self):
 	model = self.config['model']
-	usage = resp['usage']
+	# Handle both dictionary-style and attribute-style responses
+	usage = resp.usage if hasattr(resp, 'usage') else resp['usage']
+
+	# Convert usage to dictionary if it's an object
+	if hasattr(usage, 'prompt_tokens'):
+		usage = {
+			'prompt_tokens': usage.prompt_tokens,
+			'total_tokens': usage.total_tokens,
+			'completion_tokens': getattr(usage, 'completion_tokens', 0)
+		}
+
 	usage['call_cnt'] = 1
 	if 'text' in out:
 		usage['completion_chars'] = len(out['text'])
@@ -53,6 +70,7 @@ def stats_callback(out, resp, self):
 	if 'rtt' in out:
 		usage['rtt'] = out['rtt']
 		usage['rtt_cnt'] = 1
+
 	usage_stats.incr(f'usage:v4:[date]:[user]', {f'{k}:{model}':v for k,v in usage.items()})
 	usage_stats.incr(f'hourly:v4:[date]',       {f'{k}:{model}:[hour]':v for k,v in usage.items()})
 	#print('STATS_CALLBACK', usage, flush=True) # XXX

src/gui.py

@@ -248,7 +248,13 @@ def b_ask():
 					n_after=n_after,
 					model=ss['model'],
 				)
-		usage = resp.get('usage',{})
+		usage = {}
+		if hasattr(resp.get('usage'), 'prompt_tokens'):
+			usage = {
+				'prompt_tokens': resp['usage'].prompt_tokens,
+				'completion_tokens': resp['usage'].completion_tokens,
+				'total_tokens': resp['usage'].total_tokens
+			}
 		usage['cnt'] = 1
 		ss['debug']['model.query.resp'] = resp
 		ss['debug']['resp.usage'] = usage
@@ -258,7 +264,7 @@ def b_ask():
 		a = resp['text'].strip()
 		ss['answer'] = a
 		output_add(q,a)
-		st.experimental_rerun() # to enable the feedback buttons
+		st.rerun()
 
 def b_clear():
 	if st.button('clear output'):

src/model.py

@@ -38,12 +38,16 @@ def get_vectors(text_list):
 	usage = Counter()
 	for i,texts in enumerate(batch(text_list, batch_size)):
 		resp = ai.embeddings(texts)
-		v = resp['vectors']
-		u = resp['usage']
+		v = resp.get('vectors', getattr(resp, 'vectors', []))
+		u = resp.get('usage', getattr(resp, 'usage', {}))
 		u['call_cnt'] = 1
 		usage.update(u)
 		vectors.extend(v)
-	return {'vectors':vectors, 'usage':dict(usage), 'model':resp['model']}
+	return {
+		'vectors': vectors,
+		'usage': dict(usage),
+		'model': getattr(resp, 'model', resp.get('model', ''))
+	}
 
 def index_file(f, filename, fix_text=False, frag_size=0, cache=None):
 	"return vector index (dictionary) for a given PDF file"