FIX TEST Higher tolerance for AdaLoRA in test (#1897)

The test is flaky on CI, so this PR increases the tolerance to hopefully
fix the flakines. I cannot reproduce the error locally (neither on GPU
nor CPU), so I'm not 100% sure if this tolerance is enough to make the
test reliable.

tests/testing_common.py

@@ -567,6 +567,9 @@ def _test_merge_layers(self, model_id, config_cls, config_kwargs):
         atol, rtol = 1e-4, 1e-4
         if self.torch_device in ["mlu"]:
             atol, rtol = 1e-3, 1e-3  # MLU
+        if config.peft_type == "ADALORA":
+            # AdaLoRA is a bit flaky on CI, but this cannot be reproduced locally
+            atol, rtol = 1e-3, 1e-3
         if (config.peft_type == "IA3") and (model_id == "Conv2d"):
             # for some reason, the IA³ Conv2d introduces a larger error
             atol, rtol = 0.3, 0.01