Skip to content

PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度

Fix CPU CI (#1344)

Sun, Xuehao8eec5d8a20f3
diff --git a/.azure-pipelines/scripts/ut/run_ut.sh b/.azure-pipelines/scripts/ut/run_ut.sh
index 73ef823b6..974681abd 100644
--- a/.azure-pipelines/scripts/ut/run_ut.sh
+++ b/.azure-pipelines/scripts/ut/run_ut.sh
@@ -9,11 +9,11 @@ export TQDM_MININTERVAL=60
uv pip install pytest-cov pytest-html
uv pip install -r /auto-round/test/test_cpu/requirements.txt \
--extra-index-url https://download.pytorch.org/whl/cpu
+uv pip install torch==2.8.0 torchvision --index-url https://download.pytorch.org/whl/cpu
# install latest gguf for ut test
cd ~ || exit 1
git clone -b master --quiet --single-branch https://github.com/ggml-org/llama.cpp.git && cd llama.cpp/gguf-py && uv pip install . sentencepiece
-uv pip install --upgrade transformers
cd /auto-round && uv pip install .
diff --git a/.azure-pipelines/scripts/ut/run_ut_cuda.sh b/.azure-pipelines/scripts/ut/run_ut_cuda.sh
index 5ef83f527..6cc562070 100644
--- a/.azure-pipelines/scripts/ut/run_ut_cuda.sh
+++ b/.azure-pipelines/scripts/ut/run_ut_cuda.sh
@@ -95,7 +95,7 @@ function run_unit_test() {
rm -rf .coverage* *.xml *.html
uv pip install -v git+https://github.com/casper-hansen/AutoAWQ.git --no-build-isolation
- uv pip install -v git+https://github.com/ModelCloud/[email protected] --no-build-isolation
+ uv pip install https://github.com/ModelCloud/GPTQModel/releases/download/v5.6.0/gptqmodel-5.6.0+cu126torch2.9-cp310-cp310-linux_x86_64.whl --no-build-isolation
uv pip install -r https://raw.githubusercontent.com/ModelCloud/GPTQModel/refs/heads/main/requirements.txt
CMAKE_ARGS="-DGGML_CUDA=on -DLLAVA_BUILD=off" uv pip install llama-cpp-python
uv pip install 'git+https://github.com/ggml-org/llama.cpp.git#subdirectory=gguf-py'
diff --git a/auto_round/calib_dataset.py b/auto_round/calib_dataset.py
index 5776a1a1d..b4bbf65b5 100644
--- a/auto_round/calib_dataset.py
+++ b/auto_round/calib_dataset.py
@@ -456,8 +456,15 @@ def get_new_chinese_title_dataset(
@register_dataset("mbpp")
+@register_dataset("google-research-datasets/mbpp")
def get_mbpp_dataset(
- tokenizer, seqlen, dataset_name="mbpp", split=None, seed=42, apply_chat_template=False, system_prompt=None
+ tokenizer,
+ seqlen,
+ dataset_name="google-research-datasets/mbpp",
+ split=None,
+ seed=42,
+ apply_chat_template=False,
+ system_prompt=None,
):
"""Returns a dataloader for the specified dataset and split.
@@ -474,6 +481,8 @@ def get_mbpp_dataset(
"""
from datasets import load_dataset
+ dataset_name = "google-research-datasets/mbpp"
+
tokenizer_function = get_tokenizer_function(
tokenizer, seqlen, apply_chat_template=apply_chat_template, system_prompt=system_prompt
)
diff --git a/auto_round/utils/__init__.py b/auto_round/utils/__init__.py
index 898946ad3..d005b7d35 100644
--- a/auto_round/utils/__init__.py
+++ b/auto_round/utils/__init__.py
@@ -15,3 +15,28 @@
from auto_round.utils.device import *
from auto_round.utils.common import *
from auto_round.utils.model import *
+
+import transformers
+from packaging.version import Version
+
+DATASET_PATCHED = False
+# tmp batch for transformers v5.0
+if Version(transformers.__version__) >= Version("5.0.0") and not DATASET_PATCHED:
+ import datasets
+
+ datasets.original_load_dataset = datasets.load_dataset
+
+ def patch_load_dataset(*args, **kwargs):
+ for dataset_name, replace_name in [("openbookqa", "allenai/openbookqa")]:
+ if len(args) > 0 and dataset_name in args[0]:
+ args = (replace_name,) + args[1:]
+ if "path" in kwargs and kwargs["path"] is not None:
+ if dataset_name in kwargs["path"] and replace_name not in kwargs["path"]:
+ kwargs["path"] = kwargs["path"].replace(dataset_name, replace_name)
+ if "name" in kwargs and kwargs["name"] is not None:
+ if dataset_name in kwargs["name"] and replace_name not in kwargs["name"]:
+ kwargs["name"] = kwargs["name"].replace(dataset_name, replace_name)
+ return datasets.original_load_dataset(*args, **kwargs)
+
+ datasets.load_dataset = patch_load_dataset
+ DATASET_PATCHED = True
diff --git a/test/fixtures.py b/test/fixtures.py
index 01ba5cd70..624170710 100644
--- a/test/fixtures.py
+++ b/test/fixtures.py
@@ -1,6 +1,7 @@
import os
import shutil
+import datasets
import pytest
import torch
import transformers