[https://nvbugs/5474409][fix] Disable concurrent loading by default (NVIDIA#7663)

Signed-off-by: nv-guomingz <[email protected]>
Signed-off-by: Wangshanshan <[email protected]>

Author: nv-guomingz

tensorrt_llm/_torch/models/modeling_utils.py

@@ -895,7 +895,7 @@ def load_single_module(name, module):
                             p.data.copy_(module_weights[n][:])
 
     if os.environ.get("TRT_LLM_DISABLE_LOAD_WEIGHTS_IN_PARALLEL",
-                      False) in ["True", "true", "1", "yes", "y"]:
+                      True) in ["True", "true", "1", "yes", "y"]:
         for name, module in tqdm(list(model.named_modules()),
                                  desc="Loading weights"):
             load_single_module(name, module)