ADLR/megatron-lm!2148 - Add model.eval() to run_text_generation_server.py

Author: mathemakitten

tools/run_text_generation_server.py

@@ -122,6 +122,8 @@ def add_text_generate_args(parser):
 
     assert len(model) == 1, "Above condition should have caught this"
     model = model[0]
+    model.eval()
+
     if mpu.is_pipeline_first_stage() and mpu.get_tensor_model_parallel_rank() == 0:
         server = MegatronServer(model)
         server.run("0.0.0.0",port=args.port)