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 15, 08:30 UTC
[Hardware][CPU] Update intel_extension_for_pytorch 2.7.0 and move to `requirements/cpu.txt` (#18542)
Signed-off-by: Kay Yan <kay.yan@daocloud.io>
Kay Yan committed
May 23, 2025 at 19:38 UTC
7ab056c273dbf89ac1e157e539a8b2c1967fb263
3 files changed
+5
-4
docker/Dockerfile.cpu
-3
@@ -51,9 +51,6 @@ RUN --mount=type=cache,target=/root/.cache/uv \
51
uv pip install --upgrade pip && \
52
uv pip install -r requirements/cpu.txt
53
54
-RUN --mount=type=cache,target=/root/.cache/uv \
55
- uv pip install intel-openmp==2024.2.1 intel_extension_for_pytorch==2.6.0
56
-
54
ENV LD_PRELOAD="/usr/lib/x86_64-linux-gnu/libtcmalloc_minimal.so.4:/opt/venv/lib/libiomp5.so:$LD_PRELOAD"
55
56
RUN echo 'ulimit -c 0' >> ~/.bashrc
requirements/cpu.txt
+4
@@ -20,3 +20,7 @@ datasets # for benchmark scripts
20
21
# cpu cannot use triton 3.3.0
22
triton==3.2.0; platform_machine == "x86_64"
23
+
24
+# Intel Extension for PyTorch, only for x86_64 CPUs
25
+intel-openmp; platform_machine == "x86_64"
26
+intel_extension_for_pytorch==2.7.0; platform_machine == "x86_64"
vllm/model_executor/layers/quantization/ipex_quant.py
+1
-1
@@ -14,7 +14,7 @@ from vllm.model_executor.layers.quantization.base_config import (
14
from vllm.model_executor.layers.quantization.gptq import GPTQLinearMethod
15
from vllm.platforms import current_platform
16
17
-MIN_IPEX_VERSION = "2.5.0"
17
+MIN_IPEX_VERSION = "2.7.0"
18
19
20
class IPEXConfig(QuantizationConfig):