fix json mode

Signed-off-by: Praneeth Bedapudi <[email protected]>

fastdeploy/_rest.py

@@ -136,7 +136,10 @@ def on_post(self, req, resp):
 
         if success is not True:
             resp.status = falcon.HTTP_400
-            resp.data = failure_response
+            if input_type == "json":
+                resp.media = failure_response
+            else:
+                resp.data = failure_response
 
         else:
             (
@@ -146,7 +149,10 @@ def on_post(self, req, resp):
                 unique_id, is_compressed, input_type, client_timeout
             )
             resp.status = falcon.HTTP_200 if success else falcon.HTTP_400
-            resp.data = response
+            if input_type == "json":
+                resp.media = response
+            else:
+                resp.data = response
 
 
 class PrometheusMetrics(object):

recipes/text_embeddings/example.py

@@ -1,5 +1,23 @@
-example = [
-    "This is a test sentence which is a bit longer than the others and also very long than normal and some more words here and some more words here and some more there",
-    "This is another test sentence",
-    "This is a third test sentence"
-]
+# generate random sentence with words of size 1-10 characters and total 5-100 words
+
+import random
+import string
+
+words = open("words.txt", "r").read().split()
+
+def generate_random_sentence():
+    # Generate random number of words between 5-100
+    num_words = random.randint(3, 100)
+
+    sentence = []
+    for _ in range(num_words):
+        word = random.choice(words)
+        sentence.append(word)
+
+    return ' '.join(sentence)
+
+
+def example_function():
+    return [generate_random_sentence() for _ in range(random.randint(1, 10))]
+
+example = example_function()

recipes/text_embeddings/predictor.py

@@ -4,8 +4,6 @@
 
 model = SentenceTransformer('Alibaba-NLP/gte-base-en-v1.5', trust_remote_code=True, backend="onnx", model_kwargs={"file_name": "model.onnx", "provider": "CPUExecutionProvider"})
 
-from time import time
-
 def predictor(input_list, batch_size=16):
     return model.encode(input_list, convert_to_numpy=True, normalize_embeddings=True, show_progress_bar=False, batch_size=batch_size)