thking

infer/online/client.py

@@ -4,7 +4,8 @@ import re
 
 API_BASE_URL = "http://localhost:8000/v1"
 # MODEL_NAME = "/home/zwq/model/Qwen3-30B-A3B" 
-MODEL_NAME = "/home/zwq/model/Qwen3-30B-A3B-Instruct-2507"
+# MODEL_NAME = "/home/zwq/model/Qwen3-30B-A3B-Instruct-2507"
+MODEL_NAME = "/home/zwq/model/Qwen3-30B-A3B-Thinking-2507"
 
 
 class Qwen3ChatClient:
@@ -176,7 +177,7 @@ if __name__ == "__main__":
         print("--------------------------------------------------")
 
 
-    output_filename_client_all_results = './Qwen3-30B-A3B-Instruct-2507_logprobs_A800_fp16.json'
+    output_filename_client_all_results = './Qwen3-30B-A3B-Thinking-2507_logprobs_K100AI_fp16.json'
     with open(output_filename_client_all_results, 'w', encoding='utf-8') as f: 
         json.dump(results_to_save, f, indent=4, ensure_ascii=False)