⚡ add flush print in local_chat output and change default optimize yaml of...

⚡ add flush print in local_chat output and change default optimize yaml of deepseekv3 to single gpu

ktransformers/local_chat.py

@@ -32,7 +32,7 @@ custom_models = {
 ktransformer_rules_dir = os.path.dirname(os.path.abspath(__file__)) + "/optimize/optimize_rules/"
 default_optimize_rules = {
     "DeepseekV2ForCausalLM": ktransformer_rules_dir + "DeepSeek-V2-Chat.yaml",
-    "DeepseekV3ForCausalLM": ktransformer_rules_dir + "DeepSeek-V3-Chat-multi-gpu.yaml",
+    "DeepseekV3ForCausalLM": ktransformer_rules_dir + "DeepSeek-V3-Chat.yaml",
     "Qwen2MoeForCausalLM": ktransformer_rules_dir + "Qwen2-57B-A14B-Instruct.yaml",
     "LlamaForCausalLM": ktransformer_rules_dir + "Internlm2_5-7b-Chat-1m.yaml",
     "MixtralForCausalLM": ktransformer_rules_dir + "Mixtral.yaml",

ktransformers/server/backend/interfaces/transformers.py

@@ -329,14 +329,14 @@ class TransformersInterface(BackendInterfaceBase):
         self.profiler.create_and_start_timer("prefill")
         for t in self.prefill(input_ids, self.check_is_new(thread_id)):
             if t is not None:
-                print(t, end="")
+                print(t, end="",flush=True)
                 yield t
         self.profiler.pause_timer("prefill")
 
         self.profiler.create_and_start_timer("decode")
         for t in self.generate():
             if t is not None:
-                print(t, end="")
+                print(t, end="",flush=True)
                 yield t
         print("")
         self.profiler.pause_timer("decode")