Improve the artifact saving (#1904)

.github/workflows/run-max-num-tokens.yml

@@ -59,6 +59,7 @@ jobs:
         run: |
           python3 llm/run_max_num_tokens.py "${{ inputs.running_template }}"
       - name: Upload test logs
+        if: always()
         uses: actions/upload-artifact@v3
         with:
           path: tests/integration/max_num_token_results/

tests/integration/llm/run_max_num_tokens.py

@@ -17,9 +17,8 @@
     json_str = [line.decode("utf-8").strip() for line in data][0]
     model_tp_dict = json.loads(json_str)
 
-    log_id = 0
     for model_id, tp_list in model_tp_dict.items():
-        model_name = model_id
+        model_name = os.path.basename(os.path.normpath(model_id))
         print(f"Starting runs for model {model_name}:")
         if model_id[:2] == 's3':
             # download model
@@ -49,6 +48,6 @@
             model, tp, max_tokens = max_token_finder(properties)
             output = f"Summary:\nmodel: {model_name}\n tp: {tp}\n max_tokens: {max_tokens}"
             print(output)
-            with open("max_num_token_results/" + str(log_id) + "_log.txt",
+            with open(f"max_num_token_results/{model_name}_{tp}_log.txt",
                       "w") as log_file:
                 log_file.write(output)