[dependencies] Update datasets pin (#39500)

* pyarrow pin

* make fixup

* test?

* like this?

* like this?

* like this?

* datasets pin

* comment

.github/workflows/build_documentation.yml

@@ -18,6 +18,10 @@ jobs:
       notebook_folder: transformers_doc
       languages: ar de en es fr hi it ko pt tr zh ja te
       custom_container: huggingface/transformers-doc-builder
+      # Temporary pin to work around datasets exception in the docbuilder.Remove after docker images and main have
+      # the right dependencies (which **should** be the case by 2025-07-20). See
+      # https://github.com/huggingface/transformers/actions/runs/16365952006/job/46243081358?pr=38545
+      pre_command: uv pip install datasets>=2.15.0
     secrets:
       token: ${{ secrets.HUGGINGFACE_PUSH }}
       hf_token: ${{ secrets.HF_DOC_BUILD_PUSH }}

.github/workflows/build_pr_documentation.yml

@@ -15,6 +15,7 @@ jobs:
       pr_number: ${{ github.event.number }}
       package: transformers
       languages: en
-      # temporary pin to work around datasets exception in the docbuilder. See
+      # Temporary pin to work around datasets exception in the docbuilder. Remove after docker images and main have
+      # the right dependencies (which **should** be the case by 2025-07-20). See
       # https://github.com/huggingface/transformers/actions/runs/16365952006/job/46243081358?pr=38545
-      pre_command: uv pip install pyarrow==20.0.0
+      pre_command: uv pip install datasets>=2.15.0

setup.py

@@ -103,7 +103,7 @@ _deps = [
     "codecarbon>=2.8.1",
     "cookiecutter==1.7.3",
     "dataclasses",
-    "datasets!=2.5.0",
+    "datasets>=2.15.0",  # We need either this pin or pyarrow<21.0.0
     "deepspeed>=0.9.3",
     "diffusers",
     "dill<0.3.5",

src/transformers/dependency_versions_table.py

@@ -10,7 +10,7 @@ deps = {
     "codecarbon": "codecarbon>=2.8.1",
     "cookiecutter": "cookiecutter==1.7.3",
     "dataclasses": "dataclasses",
-    "datasets": "datasets!=2.5.0",
+    "datasets": "datasets>=2.15.0",
     "deepspeed": "deepspeed>=0.9.3",
     "diffusers": "diffusers",
     "dill": "dill<0.3.5",