From 424b05b840a047a7801016cf91cb19a545ea5402 Mon Sep 17 00:00:00 2001
From: Edward Li <2023edwardll@gmail.com>
Date: Wed, 11 Sep 2024 06:41:30 +0800
Subject: [PATCH] remove some warnings (#404)

---
 pyproject.toml             |  4 ++++
 src/datachain/lib/clip.py  |  2 +-
 src/datachain/lib/image.py |  2 +-
 src/datachain/lib/text.py  |  2 +-
 tests/func/test_pytorch.py | 13 ++++++++++---
 5 files changed, 17 insertions(+), 6 deletions(-)

diff --git a/pyproject.toml b/pyproject.toml
index 3f1d44aa0..222ddb0b7 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -141,6 +141,10 @@ filterwarnings = [
   "error::pytest_mock.PytestMockWarning",
   "error::pytest.PytestCollectionWarning",
   "error::sqlalchemy.exc.SADeprecationWarning",
+  "ignore::DeprecationWarning:timm.*",
+  "ignore::DeprecationWarning:botocore.auth",
+  "ignore::DeprecationWarning:datasets.utils._dill",
+  "ignore::DeprecationWarning:librosa.core.intervals",
   "ignore:Field name .* shadows an attribute in parent:UserWarning"  # datachain.lib.feature
 ]
 
diff --git a/src/datachain/lib/clip.py b/src/datachain/lib/clip.py
index 430cad6eb..9d5163739 100644
--- a/src/datachain/lib/clip.py
+++ b/src/datachain/lib/clip.py
@@ -18,7 +18,7 @@ def _get_encoder(model: Any, type: Literal["image", "text"]) -> Callable:
         hasattr(model, method_name) and inspect.ismethod(getattr(model, method_name))
     ):
         method = getattr(model, method_name)
-        return lambda x: method(torch.tensor(x))
+        return lambda x: method(torch.as_tensor(x).clone().detach())
 
     # Check for model from clip or open_clip library
     method_name = f"encode_{type}"
diff --git a/src/datachain/lib/image.py b/src/datachain/lib/image.py
index e69908f0d..317d26e17 100644
--- a/src/datachain/lib/image.py
+++ b/src/datachain/lib/image.py
@@ -34,7 +34,7 @@ def convert_image(
             from transformers.image_processing_utils import BaseImageProcessor
 
             if isinstance(transform, BaseImageProcessor):
-                img = torch.tensor(img.pixel_values[0])  # type: ignore[assignment,attr-defined]
+                img = torch.as_tensor(img.pixel_values[0]).clone().detach()  # type: ignore[assignment,attr-defined]
         except ImportError:
             pass
         if device:
diff --git a/src/datachain/lib/text.py b/src/datachain/lib/text.py
index 6f079084d..a547f170a 100644
--- a/src/datachain/lib/text.py
+++ b/src/datachain/lib/text.py
@@ -33,7 +33,7 @@ def convert_text(
         res = tokenizer(text)
 
     tokens = res.input_ids if isinstance(tokenizer, PreTrainedTokenizerBase) else res
-    tokens = torch.tensor(tokens)
+    tokens = torch.as_tensor(tokens).clone().detach()
     if device:
         tokens = tokens.to(device)
 
diff --git a/tests/func/test_pytorch.py b/tests/func/test_pytorch.py
index 29aec6688..68c6910c9 100644
--- a/tests/func/test_pytorch.py
+++ b/tests/func/test_pytorch.py
@@ -1,5 +1,6 @@
 import open_clip
 import pytest
+import torch
 from datasets import load_dataset
 from torch import Size, Tensor
 from torchvision.datasets import FakeData
@@ -33,7 +34,9 @@ def fake_dataset(catalog, fake_image_dir):
 
 
 def test_pytorch_dataset(fake_dataset):
-    transform = v2.Compose([v2.ToTensor(), v2.Resize((64, 64))])
+    transform = v2.Compose(
+        [v2.ToImage(), v2.ToDtype(torch.float32, scale=True), v2.Resize((64, 64))]
+    )
     tokenizer = open_clip.get_tokenizer("ViT-B-32")
     pt_dataset = PytorchDataset(
         name=fake_dataset.name,
@@ -49,7 +52,9 @@ def test_pytorch_dataset(fake_dataset):
 
 
 def test_pytorch_dataset_sample(fake_dataset):
-    transform = v2.Compose([v2.ToTensor(), v2.Resize((64, 64))])
+    transform = v2.Compose(
+        [v2.ToImage(), v2.ToDtype(torch.float32, scale=True), v2.Resize((64, 64))]
+    )
     pt_dataset = PytorchDataset(
         name=fake_dataset.name,
         version=fake_dataset.version,
@@ -62,7 +67,9 @@ def test_pytorch_dataset_sample(fake_dataset):
 def test_to_pytorch(fake_dataset):
     from torch.utils.data import IterableDataset
 
-    transform = v2.Compose([v2.ToTensor(), v2.Resize((64, 64))])
+    transform = v2.Compose(
+        [v2.ToImage(), v2.ToDtype(torch.float32, scale=True), v2.Resize((64, 64))]
+    )
     tokenizer = open_clip.get_tokenizer("ViT-B-32")
     pt_dataset = fake_dataset.to_pytorch(transform=transform, tokenizer=tokenizer)
     assert isinstance(pt_dataset, IterableDataset)