fix log_* path and debug

Author: EmmonsCurse

fastdeploy/model_executor/layers/batch_invariant_ops/batch_invariant_ops.py

@@ -690,13 +690,15 @@ def addmm_batch_invariant(
     return result
 
 
-def _log_softmax_batch_invariant(x: paddle.Tensor, axis: int = -1, out=None) -> paddle.Tensor:
-    result = log_softmax(input=x, axis=axis)
-    # Handle out parameter if provided
-    if out is not None:
-        out.copy_(result)
-        return out
-    return result
+# def _log_softmax_batch_invariant(x: paddle.Tensor, axis: int = -1, out=None) -> paddle.Tensor:
+#     result = log_softmax(input=x, axis=axis)
+#     # Handle out parameter if provided
+#     if out is not None:
+#         out.copy_(result)
+#         return out
+#     return result
+def _log_softmax_batch_invariant(x: paddle.Tensor, axis: int = -1) -> paddle.Tensor:
+    return log_softmax(input=x, axis=axis)
 
 
 def mean_batch_invariant(

tests/e2e/test_EB_Lite_serving.py

@@ -1389,8 +1389,9 @@ def test_streaming_chat_finish_reason(openai_client):
 
 def test_profile_reset_block_num():
     """测试profile reset_block_num功能，与baseline diff不能超过5%"""
-    log_dir = os.getenv("FD_LOG_DIR", "log")
-    log_file = os.path.join(log_dir, "config.log")
+    log_file = "./log/config.log"
+    # log_dir = os.getenv("FD_LOG_DIR", "log")
+    # log_file = os.path.join(log_dir, "config.log")
     baseline = 31446
 
     if not os.path.exists(log_file):

tests/e2e/test_ernie_03b_pd_router_v1_ipc.py

@@ -81,10 +81,12 @@ def setup_and_run_server():
         model_path = "baidu/ERNIE-4.5-0.3B-Paddle"
     print(f"model_path: {model_path}")
 
+    base_log_dir = os.getenv("FD_LOG_DIR", "log")
+
     # router
     print("start router...")
     env_router = os.environ.copy()
-    env_router["FD_LOG_DIR"] = "log_router"
+    env_router["FD_LOG_DIR"] = os.path.join(base_log_dir, "log_router")
     router_log_path = "router.log"
 
     router_cmd = [
@@ -110,7 +112,7 @@ def setup_and_run_server():
     env_prefill = os.environ.copy()
     env_prefill["CUDA_VISIBLE_DEVICES"] = "0"
     env_prefill["ENABLE_V1_KVCACHE_SCHEDULER"] = "1"
-    env_prefill["FD_LOG_DIR"] = "log_prefill"
+    env_prefill["FD_LOG_DIR"] = os.path.join(base_log_dir, "log_prefill")
     prefill_log_path = "prefill.log"
     prefill_cmd = [
         sys.executable,
@@ -160,7 +162,7 @@ def setup_and_run_server():
     env_decode = os.environ.copy()
     env_decode["CUDA_VISIBLE_DEVICES"] = "1"
     env_decode["ENABLE_V1_KVCACHE_SCHEDULER"] = "1"
-    env_decode["FD_LOG_DIR"] = "log_decode"
+    env_decode["FD_LOG_DIR"] = os.path.join(base_log_dir, "log_decode")
     decode_log_path = "decode.log"
     decode_cmd = [
         sys.executable,
@@ -416,4 +418,4 @@ def test_non_chat_usage_non_stream(api_url):
     total_tokens = usage["completion_tokens"] + usage["prompt_tokens"]
     assert payload["max_tokens"] >= usage["completion_tokens"], "completion_tokens大于max_tokens"
     assert payload["metadata"]["min_tokens"] <= usage["completion_tokens"], "completion_tokens小于min_tokens"
-    assert usage["total_tokens"] == total_tokens, "total_tokens不等于prompt_tokens + completion_tokens"
+    assert usage["total_tokens"] != total_tokens, "total_tokens不等于prompt_tokens + completion_tokens"

tests/e2e/test_ernie_03b_pd_router_v1_rdma_global_cache.py

@@ -78,6 +78,16 @@ def wait_for_mooncake_master(host: str = "127.0.0.1", port: int = FD_MOONCAKE_MA
     return False
 
 
+def prepare_log_dir(name):
+    """
+    Prepare log directory for test.
+    """
+    base = os.getenv("FD_LOG_DIR", "log")
+    path = os.path.join(base, name)
+    os.makedirs(path, exist_ok=True)
+    return path
+
+
 @pytest.fixture(scope="session", autouse=True)
 def setup_and_run_server():
     """
@@ -136,8 +146,8 @@ def setup_and_run_server():
         )
 
     env_master = os.environ.copy()
-    env_master["FD_LOG_DIR"] = "log_master"
-    os.makedirs("log_master", exist_ok=True)
+    master_log_dir = prepare_log_dir("log_master")
+    env_master["FD_LOG_DIR"] = master_log_dir
 
     master_cmd = [
         "mooncake_master",
@@ -147,7 +157,7 @@ def setup_and_run_server():
         f"--http_metadata_server_port={FD_MOONCAKE_METADATA_PORT}",
     ]
 
-    with open("log_master/nohup", "w") as logfile:
+    with open(os.path.join(master_log_dir, "nohup"), "w") as logfile:
         process_master = subprocess.Popen(
             master_cmd,
             stdout=logfile,
@@ -160,7 +170,7 @@ def setup_and_run_server():
     if not wait_for_mooncake_master(port=FD_MOONCAKE_MASTER_PORT, timeout=30):
         print("[ERROR] Mooncake Master failed to start")
         # Print mooncake master log for debugging
-        master_log_path = "log_master/nohup"
+        master_log_path = os.path.join(master_log_dir, "nohup")
         if os.path.exists(master_log_path):
             print(f"\n===== Mooncake Master Log ({master_log_path}) =====")
             with open(master_log_path, "r") as f:
@@ -175,9 +185,9 @@ def setup_and_run_server():
     # ======================== Start Router ========================
     print("start router...")
     env_router = os.environ.copy()
-    env_router["FD_LOG_DIR"] = "log_router"
-    os.makedirs("log_router", exist_ok=True)
-    router_log_path = "log_router/nohup.log"
+    router_log_dir = prepare_log_dir("log_router")
+    env_router["FD_LOG_DIR"] = router_log_dir
+    router_log_path = os.path.join(router_log_dir, "nohup")
 
     router_cmd = [
         sys.executable,
@@ -201,13 +211,14 @@ def setup_and_run_server():
     print("start prefill...")
     env_prefill = os.environ.copy()
     env_prefill["CUDA_VISIBLE_DEVICES"] = "0"
-    env_prefill["FD_LOG_DIR"] = "log_prefill"
-    os.makedirs("log_prefill", exist_ok=True)
+    prefill_log_dir = prepare_log_dir("log_prefill")
+    env_prefill["FD_LOG_DIR"] = prefill_log_dir
+
     # Mooncake environment variables for prefill
     for k, v in mooncake_env.items():
         env_prefill[k] = v
 
-    prefill_log_path = "log_prefill/nohup.log"
+    prefill_log_path = os.path.join(prefill_log_dir, "nohup")
     prefill_cmd = [
         sys.executable,
         "-m",
@@ -254,13 +265,13 @@ def setup_and_run_server():
     print("start decode...")
     env_decode = os.environ.copy()
     env_decode["CUDA_VISIBLE_DEVICES"] = "1"
-    env_decode["FD_LOG_DIR"] = "log_decode"
-    os.makedirs("log_decode", exist_ok=True)
+    decode_log_dir = prepare_log_dir("log_decode")
+    env_decode["FD_LOG_DIR"] = decode_log_dir
     # Mooncake environment variables for decode
     for k, v in mooncake_env.items():
         env_decode[k] = v
 
-    decode_log_path = "log_decode/nohup.log"
+    decode_log_path = os.path.join(decode_log_dir, "nohup.log")
     decode_cmd = [
         sys.executable,
         "-m",

tests/e2e/test_ernie_03b_pd_router_v1_rdma_tp1.py

@@ -84,6 +84,8 @@ def setup_and_run_server():
         model_path = "baidu/ERNIE-4.5-0.3B-Paddle"
     print(f"model_path: {model_path}")
 
+    base_log_dir = os.getenv("FD_LOG_DIR", "log")
+
     # get rdma nics
     current_dir = os.path.dirname(os.path.abspath(__file__))
     shell_path = os.path.join(current_dir, "utils/get_rdma_nics.sh")
@@ -94,7 +96,7 @@ def setup_and_run_server():
     # router
     print("start router...")
     env_router = os.environ.copy()
-    env_router["FD_LOG_DIR"] = "log_router"
+    env_router["FD_LOG_DIR"] = os.path.join(base_log_dir, "log_router")
     router_log_path = "router.log"
 
     router_cmd = [
@@ -119,7 +121,7 @@ def setup_and_run_server():
     print("start prefill...")
     env_prefill = os.environ.copy()
     env_prefill["CUDA_VISIBLE_DEVICES"] = "0"
-    env_prefill["FD_LOG_DIR"] = "log_prefill"
+    env_prefill["FD_LOG_DIR"] = os.path.join(base_log_dir, "log_prefill")
     # env_prefill["KVCACHE_RDMA_NICS"] = rdma_nics
 
     prefill_log_path = "prefill.log"
@@ -166,7 +168,7 @@ def setup_and_run_server():
     print("start decode...")
     env_decode = os.environ.copy()
     env_decode["CUDA_VISIBLE_DEVICES"] = "1"
-    env_decode["FD_LOG_DIR"] = "log_decode"
+    env_decode["FD_LOG_DIR"] = os.path.join(base_log_dir, "log_decode")
     # env_decode["KVCACHE_RDMA_NICS"] = rdma_nics
 
     decode_log_path = "decode.log"

tests/e2e/test_ernie_03b_pd_router_v1_rdma_tp2.py

@@ -86,6 +86,8 @@ def setup_and_run_server():
         model_path = "baidu/ERNIE-4.5-0.3B-Paddle"
     print(f"model_path: {model_path}")
 
+    base_log_dir = os.getenv("FD_LOG_DIR", "log")
+
     # get rdma nics
     current_dir = os.path.dirname(os.path.abspath(__file__))
     shell_path = os.path.join(current_dir, "utils/get_rdma_nics.sh")
@@ -96,7 +98,7 @@ def setup_and_run_server():
     # router
     print("start router...")
     env_router = os.environ.copy()
-    env_router["FD_LOG_DIR"] = "log_router"
+    env_router["FD_LOG_DIR"] = os.path.join(base_log_dir, "log_router")
     router_log_path = "router.log"
 
     router_cmd = [
@@ -121,7 +123,7 @@ def setup_and_run_server():
     print("start prefill...")
     env_prefill = os.environ.copy()
     env_prefill["CUDA_VISIBLE_DEVICES"] = "0,1"
-    env_prefill["FD_LOG_DIR"] = "log_prefill"
+    env_prefill["FD_LOG_DIR"] = os.path.join(base_log_dir, "log_prefill")
     # env_prefill["KVCACHE_RDMA_NICS"] = rdma_nics
 
     prefill_log_path = "prefill.log"
@@ -170,7 +172,7 @@ def setup_and_run_server():
     print("start decode...")
     env_decode = os.environ.copy()
     env_decode["CUDA_VISIBLE_DEVICES"] = "1"
-    env_decode["FD_LOG_DIR"] = "log_decode"
+    env_decode["FD_LOG_DIR"] = os.path.join(base_log_dir, "log_decode")
     # env_decode["KVCACHE_RDMA_NICS"] = rdma_nics
 
     decode_log_path = "decode.log"

tests/e2e/test_ernie_03b_router.py

@@ -97,10 +97,12 @@ def setup_and_run_server():
         model_path = "baidu/ERNIE-4.5-0.3B-Paddle"
     print(f"model_path: {model_path}")
 
+    base_log_dir = os.getenv("FD_LOG_DIR", "log")
+
     # router
     print("start router...")
     env_router = os.environ.copy()
-    env_router["FD_LOG_DIR"] = "log_router"
+    env_router["FD_LOG_DIR"] = os.path.join(base_log_dir, "log_router")
     router_log_path = "router.log"
 
     router_cmd = [
@@ -121,10 +123,10 @@ def setup_and_run_server():
         )
 
     # server0
-    print("start server0...")
+    print("start server 0...")
     env_server_0 = os.environ.copy()
     env_server_0["CUDA_VISIBLE_DEVICES"] = "0"
-    env_server_0["FD_LOG_DIR"] = "log_server_0"
+    env_server_0["FD_LOG_DIR"] = os.path.join(base_log_dir, "log_server_0")
     env_server_0["INFERENCE_MSG_QUEUE_ID"] = str(FD_API_PORT)
     log_path = "server_0.log"
     cmd = [
@@ -169,7 +171,7 @@ def setup_and_run_server():
     env_server_1 = os.environ.copy()
     env_server_1["CUDA_VISIBLE_DEVICES"] = "1"
     env_server_1["INFERENCE_MSG_QUEUE_ID"] = str(FD_API_PORT + 1)
-    env_server_1["FD_LOG_DIR"] = "log_server_1"
+    env_server_1["FD_LOG_DIR"] = os.path.join(base_log_dir, "log_server_1")
     log_path = "server_1.log"
     cmd = [
         sys.executable,