A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[CI/Build] upgrade compressed-tensors to 0.12.2 to address LGPLv3 (#26501)
Signed-off-by: Sangyeon Cho <josang1204@gmail.com>
Sangyeon Cho committed
Oct 14, 2025 at 01:58 UTC
a1b2d658ee229e94466f318c3f1bd28f0d35fdc5
3 files changed
+3
-3
requirements/common.txt
+1
-1
@@ -38,7 +38,7 @@ pyyaml
38
six>=1.16.0; python_version > '3.11' # transitive dependency of pandas that needs to be the latest version for python 3.12
39
setuptools>=77.0.3,<80; python_version > '3.11' # Setuptools is used by triton, we need to ensure a modern version is installed for 3.12+ so that it does not try to import distutils, which was removed in 3.12
40
einops # Required for Qwen2-VL.
41
-compressed-tensors == 0.11.0 # required for compressed-tensors
41
+compressed-tensors == 0.12.2 # required for compressed-tensors
42
depyf==0.19.0 # required for profiling and debugging with compilation config
43
cloudpickle # allows pickling lambda functions in model_executor/models/registry.py
44
watchfiles # required for http server to monitor the updates of TLS files
vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py
+1
-1
@@ -310,7 +310,7 @@ class CompressedTensorsConfig(QuantizationConfig):
310
)
311
is_float_type = (
312
weight_quant.type == QuantizationType.FLOAT
313
- and input_quant.type == QuantizationType.FLOAT.value
313
+ and input_quant.type == QuantizationType.FLOAT
314
)
315
is_4_bits = weight_quant.num_bits == 4 and input_quant.num_bits == 4
316
vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe.py
+1
-1
@@ -143,7 +143,7 @@ class CompressedTensorsMoEMethod(FusedMoEMethodBase):
143
# Prefer to use the MarlinMoE kernel when it is supported.
144
if not check_moe_marlin_supports_layer(layer, group_size):
145
if (
146
- weight_quant.strategy in QuantizationStrategy.GROUP
146
+ weight_quant.strategy == QuantizationStrategy.GROUP
147
and weight_quant.actorder
148
in (ActivationOrdering.GROUP, ActivationOrdering.DYNAMIC)
149
):