LIGHTLLM_PYNCCL_ENABLE default False (#620)

ModelTC · Nov 27, 2024 · b8ef073 · b8ef073
1 parent af0d743
commit b8ef073
Show file tree

Hide file tree

Showing 4 changed files with 4 additions and 6 deletions.
diff --git a/lightllm/common/basemodel/cuda_graph.py b/lightllm/common/basemodel/cuda_graph.py
@@ -49,7 +49,7 @@ def replay(self, input_ids, infer_state):
     @torch.no_grad()
     def warmup(self, model):
         logger.info("Begin capture cudagraph, use the --disable_cudagraph to disable it.")
-        LIGHTLLM_PYNCCL_ENABLE = os.getenv("LIGHTLLM_PYNCCL_ENABLE", "True").upper() in ["ON", "TRUE", "1"]
+        LIGHTLLM_PYNCCL_ENABLE = os.getenv("LIGHTLLM_PYNCCL_ENABLE", "False").upper() in ["ON", "TRUE", "1"]
         graph_capture_context_manager = graph_capture() if LIGHTLLM_PYNCCL_ENABLE else nullcontext()
         with graph_capture_context_manager as graph_capture_context:
             self.stream = graph_capture_context.stream if graph_capture_context is not None else None

diff --git a/lightllm/server/router/model_infer/mode_backend/base_backend.py b/lightllm/server/router/model_infer/mode_backend/base_backend.py
@@ -83,7 +83,7 @@ def init_model(self, kvargs):
         max_total_token_num = kvargs["max_total_token_num"]
 
         torch.cuda.set_device(self.tp_rank)
-        LIGHTLLM_PYNCCL_ENABLE = os.getenv("LIGHTLLM_PYNCCL_ENABLE", str(not self.disable_cudagraph)).upper() in [
+        LIGHTLLM_PYNCCL_ENABLE = os.getenv("LIGHTLLM_PYNCCL_ENABLE", "False").upper() in [
             "ON",
             "TRUE",
             "1",

diff --git a/test/model/model_infer.py b/test/model/model_infer.py
@@ -51,9 +51,8 @@ def tppart_model_infer(model_class, model_kvargs, batch_size, input_len, output_
     rank_id = model_kvargs["tp_rank"]
     world_size = model_kvargs["world_size"]
 
-    disable_cudagraph = model_kvargs.get("disable_cudagraph", False)
     torch.cuda.set_device(rank_id)
-    LIGHTLLM_PYNCCL_ENABLE = os.getenv("LIGHTLLM_PYNCCL_ENABLE", str(not disable_cudagraph)).upper() in [
+    LIGHTLLM_PYNCCL_ENABLE = os.getenv("LIGHTLLM_PYNCCL_ENABLE", "False").upper() in [
         "ON",
         "TRUE",
         "1",

diff --git a/test/model/test_settings/model_infer_batchs.py b/test/model/test_settings/model_infer_batchs.py
@@ -77,9 +77,8 @@ def tppart_model_infer(model_class, model_kvargs, batch_sizes, input_len, output
     rank_id = model_kvargs["tp_rank"]
     world_size = model_kvargs["world_size"]
 
-    disable_cudagraph = model_kvargs.get("disable_cudagraph", False)
     torch.cuda.set_device(rank_id)
-    LIGHTLLM_PYNCCL_ENABLE = os.getenv("LIGHTLLM_PYNCCL_ENABLE", str(not disable_cudagraph)).upper() in [
+    LIGHTLLM_PYNCCL_ENABLE = os.getenv("LIGHTLLM_PYNCCL_ENABLE", "False").upper() in [
         "ON",
         "TRUE",
         "1",