Had bugs in the fix that I didn't notice until using the big server

megatron/text_generation_server.py

@@ -61,9 +61,10 @@ class MegatronGenerate(Resource):
         temperature = args.temperature
         if "temperature" in request.get_json():
             temperature = request.get_json()["temperature"]
-            if not isinstance(temperature, float) or not \
-               0.0 < temperature <= 100.0:
-                return "temperature must be a positive float less than or equal to 100.0"
+            if not (type(temperature) == int or type(temperature) == float):
+                return "temperature must be a positive number less than or equal to 100.0"
+            if not (0.0 < temperature <= 100.0):
+                return "temperature must be a positive number less than or equal to 100.0"
         
         add_BOS = False
         if "add_BOS" in request.get_json():

megatron/text_generation_utils.py

@@ -185,7 +185,7 @@ def generate(model, sentences=None, tokens_to_generate=0, all_probs=False, tempe
         context_tokens_tensor, context_length_tensor = tokenize_batch(sentences, tokens_to_generate, add_BOS)
         send_generate_info(context_tokens_tensor, context_length_tensor, tokens_to_generate, all_probs, temperature)
     else:
-        context_length_tensor, context_tokens_tensor, tokens_to_generate, all_probs = receive_generate_info()
+        context_length_tensor, context_tokens_tensor, tokens_to_generate, all_probs, temperature = receive_generate_info()
 
     output = synced_generate(model, context_tokens_tensor, context_length_tensor, tokens_to_generate, all_probs, temperature)
     if output is not None: