EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

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

Browse Source 
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>
This commit is contained in:
wangxiyuan
2025-12-22 11:28:26 +08:00
committed by GitHub
parent e117b3d693
commit 492173cf89
6 changed files with 10 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm_ascend/compilation/passes/norm_quant_fusion_pass.py
View File

@@ -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:

15
vllm_ascend/compilation/passes/qknorm_rope_fusion_pass.py
View File

@@ -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()