Merge branch 'txt_seq_lens' of https://github.com/kashif/diffusers into txt_seq_lens

Author: kashif

src/diffusers/models/controlnets/controlnet_qwenimage.py

@@ -174,9 +174,9 @@ def forward(
         if txt_seq_lens is not None:
             deprecate(
                 "txt_seq_lens",
-                "0.37.0",
+                "0.39.0",
                 "Passing `txt_seq_lens` to `QwenImageControlNetModel.forward()` is deprecated and will be removed in "
-                "version 0.37.0. The text sequence length is now automatically inferred from `encoder_hidden_states` "
+                "version 0.39.0. The text sequence length is now automatically inferred from `encoder_hidden_states` "
                 "and `encoder_hidden_states_mask`.",
                 standard_warn=False,
             )
@@ -291,9 +291,9 @@ def forward(
         if txt_seq_lens is not None:
             deprecate(
                 "txt_seq_lens",
-                "0.37.0",
+                "0.39.0",
                 "Passing `txt_seq_lens` to `QwenImageMultiControlNetModel.forward()` is deprecated and will be "
-                "removed in version 0.37.0. The text sequence length is now automatically inferred from "
+                "removed in version 0.39.0. The text sequence length is now automatically inferred from "
                 "`encoder_hidden_states` and `encoder_hidden_states_mask`.",
                 standard_warn=False,
             )

src/diffusers/models/transformers/transformer_qwenimage.py

@@ -253,8 +253,8 @@ def forward(
         if txt_seq_lens is not None:
             deprecate(
                 "txt_seq_lens",
-                "0.37.0",
-                "Passing `txt_seq_lens` is deprecated and will be removed in version 0.37.0. "
+                "0.39.0",
+                "Passing `txt_seq_lens` is deprecated and will be removed in version 0.39.0. "
                 "Please use `max_txt_seq_len` instead. "
                 "The new parameter accepts a single int or tensor value representing the maximum text sequence length.",
                 standard_warn=False,
@@ -907,8 +907,8 @@ def forward(
         if txt_seq_lens is not None:
             deprecate(
                 "txt_seq_lens",
-                "0.37.0",
-                "Passing `txt_seq_lens` is deprecated and will be removed in version 0.37.0. "
+                "0.39.0",
+                "Passing `txt_seq_lens` is deprecated and will be removed in version 0.39.0. "
                 "Please use `encoder_hidden_states_mask` instead. "
                 "The mask-based approach is more flexible and supports variable-length sequences.",
                 standard_warn=False,

tests/models/transformers/test_models_transformer_qwenimage.py

@@ -145,6 +145,23 @@ def test_infers_text_seq_len_from_mask(self):
         self.assertIsNone(per_sample_len_none)
         self.assertIsNone(normalized_mask_none)
 
+    def test_deprecated_txt_seq_lens_warning(self):
+        """Test that passing the deprecated txt_seq_lens parameter raises a FutureWarning."""
+        init_dict, inputs = self.prepare_init_args_and_inputs_for_common()
+        model = self.model_class(**init_dict).to(torch_device)
+        model.eval()
+
+        # Add the deprecated txt_seq_lens parameter
+        inputs["txt_seq_lens"] = [inputs["encoder_hidden_states"].shape[1]]
+
+        with self.assertWarns(FutureWarning) as warning:
+            with torch.no_grad():
+                _ = model(**inputs)
+
+        # Verify the warning message mentions the deprecated parameter
+        self.assertIn("txt_seq_lens", str(warning.warning))
+        self.assertIn("deprecated", str(warning.warning).lower())
+
     def test_non_contiguous_attention_mask(self):
         """Test that non-contiguous masks work correctly (e.g., [1, 0, 1, 0, 1, 0, 0])"""
         init_dict, inputs = self.prepare_init_args_and_inputs_for_common()