biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix] Add proper comparison for package versions (#22314)

Browse Source 
Signed-off-by: Syed Muhammad Bin Asif <syedmba7@connect.hku.hk>
This commit is contained in:
Syed Muhammad Bin Asif
2025-08-07 11:31:03 +08:00
committed by GitHub
parent 5e9455ae8f
commit 609b533cb6
13 changed files with 40 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/layers/quantization/bitblas.py
View File

@@ -3,6 +3,7 @@
from typing import Any, Optional
import torch
from packaging import version
from vllm.logger import init_logger
from vllm.model_executor.layers.linear import LinearBase, LinearMethodBase
@@ -45,7 +46,8 @@ class BitBLASConfig(QuantizationConfig):
) -> None:
try:
import bitblas
if bitblas.__version__ < MINIMUM_BITBLAS_VERSION:
if version.parse(bitblas.__version__) < version.parse(
MINIMUM_BITBLAS_VERSION):
raise ImportError(
"bitblas version is wrong. Please "
f"install bitblas>={MINIMUM_BITBLAS_VERSION}")

7
vllm/model_executor/layers/quantization/bitsandbytes.py
View File

@@ -4,6 +4,7 @@
from typing import Any, Callable, Optional, Union
import torch
from packaging import version
from vllm.model_executor.layers.fused_moe.layer import (FusedMoE,
FusedMoEMethodBase)
@@ -169,7 +170,8 @@ class BitsAndBytesLinearMethod(LinearMethodBase):
def __init__(self, quant_config: BitsAndBytesConfig):
try:
import bitsandbytes
if bitsandbytes.__version__ < "0.46.1":
if version.parse(
bitsandbytes.__version__) < version.parse("0.46.1"):
raise ImportError("bitsandbytes version is wrong. Please "
"install bitsandbytes>=0.46.1.")
except ImportError as err:
@@ -412,7 +414,8 @@ class BitsAndBytesMoEMethod(FusedMoEMethodBase):
def __init__(self, quant_config: BitsAndBytesConfig):
try:
import bitsandbytes
if bitsandbytes.__version__ < "0.46.1":
if version.parse(
bitsandbytes.__version__) < version.parse("0.46.1"):
raise ImportError("bitsandbytes version is wrong. Please "
"install bitsandbytes>=0.46.1.")
except ImportError as err:

3
vllm/model_executor/layers/quantization/deepspeedfp.py
View File

@@ -6,6 +6,7 @@ from typing import Any, Optional
import torch
import torch.nn as nn
import torch.nn.functional as F
from packaging import version
from vllm.model_executor.layers.linear import LinearBase, LinearMethodBase
from vllm.model_executor.layers.quantization import QuantizationMethods
@@ -145,7 +146,7 @@ class DeepSpeedFPParameter(nn.Parameter):
quant_config: DeepSpeedFPConfig):
try:
import deepspeed
if deepspeed.__version__ < "0.14.2":
if version.parse(deepspeed.__version__) < version.parse("0.14.2"):
raise ImportError("deepspeed version is wrong. Please "
"install deepspeed>=0.14.2.")
from deepspeed.ops.fp_quantizer import FP_Quantize

4
vllm/model_executor/layers/quantization/gptq_bitblas.py
View File

@@ -3,6 +3,7 @@
from typing import Any, Optional
import torch
from packaging import version
from torch.nn.parameter import Parameter
from vllm.logger import init_logger
@@ -63,7 +64,8 @@ class GPTQBitBLASConfig(QuantizationConfig):
try:
import bitblas
if bitblas.__version__ < MINIMUM_BITBLAS_VERSION:
if version.parse(bitblas.__version__) < version.parse(
MINIMUM_BITBLAS_VERSION):
raise ImportError(
"bitblas version is wrong. Please "
f"install bitblas>={MINIMUM_BITBLAS_VERSION}")

7
vllm/model_executor/layers/quantization/ipex_quant.py
View File

@@ -4,6 +4,7 @@
from typing import Any, Optional
import torch
from packaging import version
from vllm.model_executor.layers.linear import (LinearBase, LinearMethodBase,
UnquantizedLinearMethod)
@@ -135,7 +136,8 @@ class IPEXGPTQLinearMethod(GPTQLinearMethod):
try:
import intel_extension_for_pytorch as ipex
if ipex.__version__ < MIN_IPEX_VERSION:
if version.parse(
ipex.__version__) < version.parse(MIN_IPEX_VERSION):
raise ImportError(
"intel_extension_for_pytorch version is "
"wrong. Please install "
@@ -199,7 +201,8 @@ class IPEXAWQLinearMethod(AWQLinearMethod):
try:
import intel_extension_for_pytorch as ipex
if ipex.__version__ < MIN_IPEX_VERSION:
if version.parse(
ipex.__version__) < version.parse(MIN_IPEX_VERSION):
raise ImportError(
"intel_extension_for_pytorch version is "
"wrong. Please install "

4
vllm/model_executor/layers/quantization/kernels/mixed_precision/bitblas.py
View File

@@ -4,6 +4,7 @@
from typing import Optional
import torch
from packaging import version
from vllm.logger import init_logger
from vllm.model_executor.layers.quantization.base_config import (
@@ -110,7 +111,8 @@ class BitBLASLinearKernel(MPLinearKernel):
try:
import bitblas
if bitblas.__version__ < MINIMUM_BITBLAS_VERSION:
if version.parse(bitblas.__version__) < version.parse(
MINIMUM_BITBLAS_VERSION):
raise ImportError(
"bitblas version is wrong. Please "
f"install bitblas>={MINIMUM_BITBLAS_VERSION}")

4
vllm/model_executor/layers/quantization/utils/bitblas_utils.py
View File

@@ -3,6 +3,7 @@
from typing import Optional
import torch
from packaging import version
from vllm.platforms import current_platform
from vllm.scalar_type import ScalarType, scalar_types
@@ -75,7 +76,8 @@ def _check_bitblas_supported(
# Finally, check if bitblas is installed
try:
import bitblas
if bitblas.__version__ < MINIMUM_BITBLAS_VERSION:
if version.parse(
bitblas.__version__) < version.parse(MINIMUM_BITBLAS_VERSION):
raise ImportError("bitblas version is wrong. Please "
f"install bitblas>={MINIMUM_BITBLAS_VERSION}")
except ImportError:

5
vllm/model_executor/layers/quantization/utils/w8a8_utils.py
View File

@@ -4,6 +4,7 @@
from typing import Callable, Optional, Union
import torch
from packaging import version
from vllm import _custom_ops as ops
from vllm import envs
@@ -21,8 +22,8 @@ TORCH_DEVICE_IDENTITY = None
# torch._scaled_mm rowwise feature.
# The condition is determined once as the operations
# are time consuming.
USE_ROWWISE_TORCH_SCALED_MM = (current_platform.is_rocm()
and torch.__version__[0:3] >= "2.7"
USE_ROWWISE_TORCH_SCALED_MM = (current_platform.is_rocm() and version.parse(
torch.__version__) >= version.parse("2.7")
and current_platform.has_device_capability(94))