Fix hub offline (#680)

* set default cache dir * added offline test * fix * finish
huggingface · Apr 23, 2024 · 00581ab · 00581ab
1 parent 7db4c7e
commit 00581ab
Show file tree

Hide file tree

Showing 14 changed files with 77 additions and 22 deletions.
diff --git a/.github/workflows/test_offline.yaml b/.github/workflows/test_offline.yaml
@@ -0,0 +1,40 @@
+name: Offline usage / Python - Test
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+jobs:
+  build:
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: [3.9]
+        os: [ubuntu-latest]
+
+    runs-on: ${{ matrix.os }}
+    steps:
+      - uses: actions/checkout@v3
+      - name: Setup Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v3
+        with:
+          python-version: ${{ matrix.python-version }}
+      - name: Install dependencies
+        run: |
+          pip install .[tests,openvino]
+      - name: Test
+        run: |
+          HF_HOME=/tmp/ huggingface-cli download hf-internal-testing/tiny-random-gpt2
+          HF_HOME=/tmp/ HF_HUB_OFFLINE=1 optimum-cli export openvino --model hf-internal-testing/tiny-random-gpt2 gpt2_openvino --task text-generation
+
+          huggingface-cli download hf-internal-testing/tiny-random-gpt2
+          HF_HUB_OFFLINE=1 optimum-cli export openvino --model hf-internal-testing/tiny-random-gpt2 gpt2_openvino --task text-generation
+
+          pytest tests/openvino/test_modeling.py -k "test_load_from_hub" -s -vvvvv
+          HF_HUB_OFFLINE=1 pytest tests/openvino/test_modeling.py -k "test_load_from_hub" -s -vvvvv
diff --git a/optimum/commands/export/openvino.py b/optimum/commands/export/openvino.py
@@ -18,6 +18,8 @@
 from pathlib import Path
 from typing import TYPE_CHECKING, Optional
 
+from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE
+
 from ...exporters import TasksManager
 from ...intel.utils.import_utils import DIFFUSERS_IMPORT_ERROR, is_diffusers_available
 from ..base import BaseOptimumCLICommand, CommandInfo
@@ -47,7 +49,9 @@ def parse_args_openvino(parser: "ArgumentParser"):
             f" {str(TasksManager.get_all_tasks())}. For decoder models, use `xxx-with-past` to export the model using past key values in the decoder."
         ),
     )
-    optional_group.add_argument("--cache_dir", type=str, default=None, help="Path indicating where to store cache.")
+    optional_group.add_argument(
+        "--cache_dir", type=str, default=HUGGINGFACE_HUB_CACHE, help="Path indicating where to store cache."
+    )
     optional_group.add_argument(
         "--framework",
         type=str,

diff --git a/optimum/exporters/openvino/__main__.py b/optimum/exporters/openvino/__main__.py
@@ -16,6 +16,7 @@
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, Callable, Dict, Optional, Union
 
+from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE
 from requests.exceptions import ConnectionError as RequestsConnectionError
 from transformers import AutoConfig, AutoTokenizer, PreTrainedTokenizerBase
 
@@ -48,7 +49,7 @@ def main_export(
     task: str = "auto",
     device: str = "cpu",
     framework: Optional[str] = None,
-    cache_dir: Optional[str] = None,
+    cache_dir: str = HUGGINGFACE_HUB_CACHE,
     trust_remote_code: bool = False,
     pad_token_id: Optional[int] = None,
     subfolder: str = "",

diff --git a/optimum/intel/generation/modeling.py b/optimum/intel/generation/modeling.py
@@ -21,6 +21,7 @@
 
 import torch
 from huggingface_hub import hf_hub_download
+from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE
 from transformers import AutoConfig, AutoModelForCausalLM, GenerationConfig, PretrainedConfig, PreTrainedModel
 from transformers.generation import GenerationMixin
 from transformers.modeling_outputs import CausalLMOutputWithPast
@@ -356,7 +357,7 @@ def _from_pretrained(
         use_auth_token: Optional[Union[bool, str, None]] = None,
         revision: Optional[Union[str, None]] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         file_name: Optional[str] = WEIGHTS_NAME,
         local_files_only: bool = False,
         use_cache: bool = True,
@@ -400,7 +401,7 @@ def _from_transformers(
         use_auth_token: Optional[Union[bool, str]] = None,
         revision: Optional[str] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         subfolder: str = "",
         local_files_only: bool = False,
         use_cache: bool = True,

diff --git a/optimum/intel/ipex/modeling_base.py b/optimum/intel/ipex/modeling_base.py
@@ -22,6 +22,7 @@
 import intel_extension_for_pytorch as ipex
 import torch
 from huggingface_hub import hf_hub_download
+from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE
 from intel_extension_for_pytorch.cpu._auto_kernel_selection import _enable_tpp
 from intel_extension_for_pytorch.transformers.optimize import get_dummy_input
 from transformers import (
@@ -153,7 +154,7 @@ def _from_transformers(
         use_auth_token: Optional[Union[bool, str]] = None,
         revision: Optional[str] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         subfolder: str = "",
         local_files_only: bool = False,
         torch_dtype: Optional[Union[str, "torch.dtype"]] = None,
@@ -190,7 +191,7 @@ def _from_pretrained(
         use_auth_token: Optional[Union[bool, str, None]] = None,
         revision: Optional[Union[str, None]] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         file_name: Optional[str] = WEIGHTS_NAME,
         local_files_only: bool = False,
         subfolder: str = "",

diff --git a/optimum/intel/neural_compressor/modeling_base.py b/optimum/intel/neural_compressor/modeling_base.py
@@ -20,6 +20,7 @@
 
 import torch
 from huggingface_hub import hf_hub_download
+from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE
 from neural_compressor.utils.pytorch import load
 from transformers import (
     AutoConfig,
@@ -104,7 +105,7 @@ def _from_pretrained(
         use_auth_token: Optional[Union[bool, str, None]] = None,
         revision: Optional[Union[str, None]] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         file_name: str = WEIGHTS_NAME,
         local_files_only: bool = False,
         subfolder: str = "",

diff --git a/optimum/intel/openvino/loaders.py b/optimum/intel/openvino/loaders.py
@@ -25,7 +25,7 @@
     import safetensors
 
 import openvino
-from huggingface_hub.constants import HF_HUB_OFFLINE
+from huggingface_hub.constants import HF_HUB_OFFLINE, HUGGINGFACE_HUB_CACHE
 from openvino.runtime import Type
 from openvino.runtime import opset11 as ops
 from openvino.runtime.passes import Manager, Matcher, MatcherPass, WrapType
@@ -37,7 +37,7 @@
 try:
     from diffusers.utils import DIFFUSERS_CACHE
 except ImportError:
-    DIFFUSERS_CACHE = None
+    DIFFUSERS_CACHE = HUGGINGFACE_HUB_CACHE
 
 
 logger = logging.getLogger(__name__)

diff --git a/optimum/intel/openvino/modeling.py b/optimum/intel/openvino/modeling.py
@@ -23,6 +23,7 @@
 import torch
 import transformers
 from huggingface_hub import model_info
+from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE
 from transformers import (
     AutoConfig,
     AutoModel,
@@ -423,7 +424,7 @@ def _from_transformers(
         use_auth_token: Optional[Union[bool, str]] = None,
         revision: Optional[str] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         subfolder: str = "",
         local_files_only: bool = False,
         task: Optional[str] = None,
@@ -585,7 +586,7 @@ def from_pretrained(
         use_auth_token: Optional[Union[bool, str]] = None,
         revision: Optional[str] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         subfolder: str = "",
         local_files_only: bool = False,
         task: Optional[str] = None,

diff --git a/optimum/intel/openvino/modeling_base.py b/optimum/intel/openvino/modeling_base.py
@@ -20,6 +20,7 @@
 
 import openvino
 from huggingface_hub import hf_hub_download
+from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE
 from openvino import Core, convert_model
 from openvino._offline_transformations import apply_moc_transformations, compress_model_transformation
 from transformers import GenerationConfig, PretrainedConfig
@@ -171,7 +172,7 @@ def _from_pretrained(
         use_auth_token: Optional[Union[bool, str, None]] = None,
         revision: Optional[Union[str, None]] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         file_name: Optional[str] = None,
         subfolder: str = "",
         from_onnx: bool = False,
@@ -300,7 +301,7 @@ def _from_transformers(
         use_auth_token: Optional[Union[bool, str]] = None,
         revision: Optional[str] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         subfolder: str = "",
         local_files_only: bool = False,
         task: Optional[str] = None,
@@ -368,7 +369,7 @@ def _to_load(
         use_auth_token: Optional[Union[bool, str]] = None,
         revision: Optional[str] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         local_files_only: bool = False,
         stateful: bool = False,
         **kwargs,

diff --git a/optimum/intel/openvino/modeling_base_seq2seq.py b/optimum/intel/openvino/modeling_base_seq2seq.py
@@ -20,6 +20,7 @@
 
 import openvino
 from huggingface_hub import hf_hub_download
+from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE
 from openvino._offline_transformations import apply_moc_transformations, compress_model_transformation
 from transformers import GenerationConfig, PretrainedConfig
 from transformers.file_utils import add_start_docstrings
@@ -111,7 +112,7 @@ def _from_pretrained(
         use_auth_token: Optional[Union[bool, str]] = None,
         revision: Optional[str] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         encoder_file_name: Optional[str] = None,
         decoder_file_name: Optional[str] = None,
         decoder_with_past_file_name: Optional[str] = None,
@@ -222,7 +223,7 @@ def _from_transformers(
         use_auth_token: Optional[Union[bool, str]] = None,
         revision: Optional[str] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         subfolder: str = "",
         local_files_only: bool = False,
         task: Optional[str] = None,

diff --git a/optimum/intel/openvino/modeling_decoder.py b/optimum/intel/openvino/modeling_decoder.py
@@ -21,6 +21,7 @@
 import numpy as np
 import openvino
 import torch
+from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE
 from openvino.preprocess import PrePostProcessor
 from openvino.runtime import Core, Tensor, Type
 from transformers import AutoModelForCausalLM, AutoTokenizer, PretrainedConfig
@@ -221,7 +222,7 @@ def _from_transformers(
         use_auth_token: Optional[Union[bool, str]] = None,
         revision: Optional[str] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         subfolder: str = "",
         local_files_only: bool = False,
         task: Optional[str] = None,
@@ -565,7 +566,7 @@ def _from_pretrained(
         use_auth_token: Optional[Union[bool, str, None]] = None,
         revision: Optional[Union[str, None]] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         file_name: Optional[str] = None,
         subfolder: str = "",
         from_onnx: bool = False,

diff --git a/optimum/intel/openvino/modeling_diffusion.py b/optimum/intel/openvino/modeling_diffusion.py
@@ -35,6 +35,7 @@
 from diffusers.schedulers.scheduling_utils import SCHEDULER_CONFIG_NAME
 from diffusers.utils import CONFIG_NAME, is_invisible_watermark_available
 from huggingface_hub import snapshot_download
+from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE
 from openvino._offline_transformations import compress_model_transformation
 from openvino.runtime import Core
 from transformers import CLIPFeatureExtractor, CLIPTokenizer
@@ -208,7 +209,7 @@ def _from_pretrained(
         config: Dict[str, Any],
         use_auth_token: Optional[Union[bool, str]] = None,
         revision: Optional[str] = None,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         vae_decoder_file_name: Optional[str] = None,
         text_encoder_file_name: Optional[str] = None,
         unet_file_name: Optional[str] = None,
@@ -400,7 +401,7 @@ def _from_transformers(
         use_auth_token: Optional[Union[bool, str]] = None,
         revision: Optional[str] = None,
         force_download: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         local_files_only: bool = False,
         tokenizer: Optional["CLIPTokenizer"] = None,
         scheduler: Union["DDIMScheduler", "PNDMScheduler", "LMSDiscreteScheduler"] = None,

diff --git a/optimum/intel/openvino/modeling_timm.py b/optimum/intel/openvino/modeling_timm.py
@@ -19,6 +19,7 @@
 import numpy as np
 import timm
 import torch
+from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE
 from packaging import version
 from timm.layers.config import set_fused_attn
 from timm.models._hub import load_model_config_from_hf
@@ -55,7 +56,7 @@ class TimmConfig(PretrainedConfig):
     def from_pretrained(
         cls,
         pretrained_model_name_or_path: Union[str, os.PathLike],
-        cache_dir: Optional[Union[str, os.PathLike]] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
         force_download: bool = False,
         local_files_only: bool = False,
         token: Optional[Union[str, bool]] = None,

diff --git a/optimum/intel/openvino/quantization.py b/optimum/intel/openvino/quantization.py
@@ -25,6 +25,7 @@
 import openvino
 import torch
 import transformers
+from huggingface_hub.constants import HUGGINGFACE_HUB_CACHE
 from nncf import CompressWeightsMode, SensitivityMetric
 from nncf.quantization.advanced_parameters import AdvancedSmoothQuantParameters, OverflowFix
 from nncf.torch import register_module
@@ -541,7 +542,7 @@ def get_calibration_dataset(
         preprocess_function: Optional[Callable] = None,
         preprocess_batch: bool = True,
         use_auth_token: bool = False,
-        cache_dir: Optional[str] = None,
+        cache_dir: str = HUGGINGFACE_HUB_CACHE,
     ) -> datasets.Dataset:
         """
         Create the calibration `datasets.Dataset` to use for the post-training static quantization calibration step.