build(deps): update transformers requirement from <4.52,>=4.51.3 to >=4.51.3,<4.54 (#2084)

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Jirka B <j.borovec+github@gmail.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Jirka Borovec <6035284+Borda@users.noreply.github.com>

Author: 3 people

.github/workflows/cpu-tests.yml

@@ -80,8 +80,8 @@ jobs:
         os: ["ubuntu-22.04"]
         python-version: ["3.9", "3.10", "3.11", "3.12", "3.13"]
         include:
-          - { os: "macOS-14", python-version: "3.9" }
-          - { os: "windows-2022", python-version: "3.9" }
+          - { os: "macOS-14", python-version: "3.10" }
+          - { os: "windows-2022", python-version: "3.10" }
     timeout-minutes: 35
     steps:
       - name: Checkout generic

pyproject.toml

@@ -65,7 +65,7 @@ optional-dependencies.extra = [
   # litgpt.pretrain:
   "tensorboard>=2.14",
   "torchmetrics>=1.3.1",
-  "transformers>=4.51.3,<4.52",
+  "transformers>=4.51.3,<4.54",
   # litdata, only on non-Windows:
   "uvloop>=0.2; sys_platform!='win32'",
   # litgpt.data.prepare_slimpajama.py:

tests/test_adapter.py

@@ -359,7 +359,7 @@ def test_against_original_gemma_2(model_name, device, dtype):
     assert x.size(1) == T
     ours_y = ours_model(x)
     theirs_y = theirs_model(x)["logits"].to(dtype)  # HF converts logits to float
-    torch.testing.assert_close(ours_y, theirs_y)
+    torch.testing.assert_close(ours_y, theirs_y, atol=1e-4, rtol=1e-5)
 
 
 @torch.inference_mode()
@@ -430,7 +430,7 @@ def test_against_original_gemma_3(model_name, device, dtype):
     assert x.size(1) == T
     ours_y = ours_model(x)
     theirs_y = theirs_model(x)["logits"].to(dtype)  # HF converts logits to float
-    torch.testing.assert_close(ours_y, theirs_y)
+    torch.testing.assert_close(ours_y, theirs_y, atol=1e-4, rtol=1e-5)
 
 
 def test_load_legacy_state_dict():

tests/test_adapter_v2.py

@@ -313,8 +313,12 @@ def test_against_original_gemma_2(model_name):
     ours_y = ours_model(x)
     theirs_y = theirs_model(x)["logits"].to(dtype)  # HF converts logits to float
     torch.testing.assert_close(
-        ours_y, theirs_y, rtol=3e-5, atol=3e-5
-    )  # some macOS devices have numerical differences, hence the tol bump
+        # some macOS devices have numerical differences, hence the tol bump
+        ours_y,
+        theirs_y,
+        atol=1e-4,
+        rtol=1e-5,
+    )
 
 
 @torch.inference_mode()

tests/test_lora.py

@@ -685,10 +685,11 @@ def test_against_original_gemma_2(model_name):
     assert x.size(1) == T
     ours_y = ours_model(x)
     theirs_y = theirs_model(x)["logits"].to(dtype)  # HF converts logits to float
-    torch.testing.assert_close(ours_y, theirs_y, rtol=3e-5, atol=3e-5)
+    torch.testing.assert_close(ours_y, theirs_y, atol=1e-4, rtol=1e-5)
 
 
 @torch.inference_mode()
+@pytest.mark.flaky(reruns=3)
 @pytest.mark.parametrize("model_name", ("gemma-3-1b-it", "gemma-3-4b-it", "gemma-3-12b-it", "gemma-3-27b-it"))
 def test_against_original_gemma_3(model_name):
     device = torch.device("cpu")