[Misc] Cleanup useless print and logger (#5220)

1. Remove useless print
2. use vLLM logger
3. change useless INFO to DEBUG level

- vLLM version: release/v0.13.0
- vLLM main:
ad32e3e19c

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

examples/eplb/eplb_strategy.py

@@ -1,7 +1,6 @@
 # coding=utf-8
 # Copyright (c) Huawei Technologies Co., Ltd. 2025-2025. All rights reserved.
 import json
-import logging
 import os
 
 import matplotlib.pyplot as plt  # type: ignore
@@ -11,8 +10,6 @@ import torch
 os.environ["VLLM_USE_MODELSCOPE"] = "True"
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
 
-logger = logging.getLogger("msit_logger")
-
 
 def save_matrix_to_json(output_path, file_name, deployment):
     num_layers = deployment.shape[0]

vllm_ascend/compilation/passes/norm_quant_fusion_pass.py

@@ -15,13 +15,12 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 #
-import logging
-
 import torch
 import torch._inductor.pattern_matcher as pm
 from torch._inductor.pattern_matcher import PatternMatcherPass
 from vllm.compilation.vllm_inductor_pass import VllmInductorPass
 from vllm.config import VllmConfig
+from vllm.logger import logger
 
 
 class AddRMSNormQuantPattern:
@@ -288,7 +287,7 @@ class AddRMSNormQuantFusionPass(VllmInductorPass):
 
         dtype = vllm_config.model_config.dtype
         if dtype not in (torch.bfloat16, torch.float16):
-            logging.info("Quant fusion not enabled: unsupported dtype %s",
+            logger.debug("Quant fusion not enabled: unsupported dtype %s",
                          dtype)
             return
 
@@ -306,7 +305,7 @@ class AddRMSNormQuantFusionPass(VllmInductorPass):
     def __call__(self, graph: torch.fx.Graph):
         self.begin()
         self.matched_count = self.pattern_match_passes.apply(graph)
-        logging.debug("Replaced %s patterns", self.matched_count)
+        logger.debug("Replaced %s patterns", self.matched_count)
         self.end_and_log()
 
     def is_applicable(self, runtime_shape: int | None = None) -> bool:

vllm_ascend/compilation/passes/qknorm_rope_fusion_pass.py

@@ -15,8 +15,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 #
-import logging
-
 import torch
 import torch._inductor.pattern_matcher as pm
 from torch._inductor.pattern_matcher import (PatternMatcherPass,
@@ -24,6 +22,7 @@ from torch._inductor.pattern_matcher import (PatternMatcherPass,
 from vllm.attention.layer import Attention
 from vllm.compilation.vllm_inductor_pass import VllmInductorPass
 from vllm.config import VllmConfig, get_layers_from_vllm_config
+from vllm.logger import logger
 
 
 class QKNormRopeFusionPattern:
@@ -237,7 +236,7 @@ class QKNormRopeFusionPass(VllmInductorPass):
 
         dtype = vllm_config.model_config.dtype
         if dtype not in (torch.bfloat16, torch.float16):
-            logging.info(
+            logger.debug(
                 "QKNorm and Rope fusion not enabled: unsupported dtype %s",
                 dtype)
             return
@@ -246,14 +245,14 @@ class QKNormRopeFusionPass(VllmInductorPass):
         attn_layers: dict[str, Attention] = get_layers_from_vllm_config(
             vllm_config, Attention)
         if len(attn_layers) == 0:
-            logging.info(
+            logger.debug(
                 "QKNorm and Rope fusion enabled, but no Attention layers were discovered."
             )
             return
         layer = next(iter(attn_layers.values()))
         for epsilon in [1e-6, 1e-5]:
             if layer.head_size != 128:
-                logging.debug(
+                logger.debug(
                     "QKNorm and Rope fusion not enabled: head_dim %d is not equal of 128",
                     layer.head_size)
                 continue
@@ -274,13 +273,13 @@ class QKNormRopeFusionPass(VllmInductorPass):
     def __call__(self, graph: torch.fx.Graph):
         self.begin()
         self.matched_count = self.pattern_match_passes.apply(graph)
-        logging.debug("Fused %s QKNorm and Rope patterns", self.matched_count)
-        logging.debug("Patterns registered for replacement:")
+        logger.debug("Fused %s QKNorm and Rope patterns", self.matched_count)
+        logger.debug("Patterns registered for replacement:")
         pattern_idx = 0
         for pattern_entry in self.pattern_match_passes.patterns.values():
             for p in pattern_entry:
                 p_str = PatternPrettyPrinter.run(p.pattern)
-                logging.debug("Pattern %d: %s", pattern_idx, p_str)
+                logger.debug("Pattern %d: %s", pattern_idx, p_str)
                 pattern_idx += 1
         self.end_and_log()
 

vllm_ascend/eplb/core/policy/policy_dynamic_ep_v2.py

@@ -202,7 +202,6 @@ class DynamicEplbV2(EplbPolicy):
             for index, target_weight in enumerate(sorted_weights):
                 expert_id, original_weight = target_weight
                 if original_weight == -1:
-                    print("Error:Redundant expert failure re-occurred")
                     redundancy_successful = True
                     break
                 redundancy_successful = False
@@ -712,7 +711,6 @@ class DynamicEplbV2(EplbPolicy):
         max_heat_per_layer_after = np.zeros([layer_num])
         sum_num = 0
         for layer in range(layer_num):
-            # print(f"Load imbalance ratio of layer {layer} under the new workload", layer_initial_imbalance[layer])
             if layer_initial_imbalance[layer] < 1.01:
                 global_deployment[layer] = info.placement_table[layer]
                 continue
@@ -734,13 +732,11 @@ class DynamicEplbV2(EplbPolicy):
                 layer_workloads[layer], info.placement_table[layer],
                 expert_from_device[layer], num_node, is_node_redundant,
                 rendun_pos)
-            # print(layer, f"Imbalance Ratio after Redundancy Adjustment:", self.safe_divide(max_workload, ave_workload))
 
             global_deployment[layer], new_max_workload = self.exchange_experts(
                 result, com_between_devices, num_node, num_npus,
                 is_node_redundant, ave_workload, increment,
                 num_redundancy_expert, info.placement_table[layer])
-            # print(layer, f"Imbalance Ratio after Swap Adjustment:", self.safe_divide(new_max_workload, ave_workload))
 
             for device_id in range(num_npus):
                 com_between_devices[device_id] = {

vllm_ascend/eplb/core/policy/policy_flashlb.py

@@ -411,7 +411,6 @@ class FlashLB(EplbPolicy):
     def compute_rank_load(self, deployment: np.ndarray, hotness: np.ndarray):
         n_stage, N = hotness.shape
         if np.any(deployment < 0):
-            print(f"Invalid deployment with negative values: {deployment}")
             raise ValueError("Deployment table contains negative values.")
         counts = np.bincount(deployment.reshape(-1), minlength=N)
         unit_hotness = np.divide(hotness,
@@ -504,8 +503,6 @@ class FlashLB(EplbPolicy):
             stage_weights,
             recorsive=False,
         )
-        if np.any(new_deployment < 0):
-            print(f"{new_deployment=}")
         new_par = self.compute_rank_load(new_deployment, hotness)
 
         return new_deployment, new_par, current_par

vllm_ascend/utils.py

@@ -1007,7 +1007,6 @@ def get_flashcomm2_config_and_validate(ascend_config, vllm_config):
 
     if not flashcomm2_enable():
         flashcomm2_oproj_shared = False
-        logger.info("FLASHCOMM2 not enable.")
         return flashcomm2_oproj_tp_size, flashcomm2_oproj_shared
 
     logger.info(