convert : fix DeepseekV4 rope parameters with transformers 5.x (#26673)

diff --git a/conversion/deepseek.py b/conversion/deepseek.py
index 3dca369cf9f516a4b1bd0374dee34d0f81093bbb..1846ca4010ec1bf5e5af74f5d0da9eb7735081a8 100644 (file)
--- a/conversion/deepseek.py
+++ b/conversion/deepseek.py
@@ -533,6 +533,13 @@ class DeepseekV4Model(TextModel):
         for key, value in raw_hparams.items():
             self.hparams.setdefault(key, value)
 
+        # workaround for special rope_parameters (main/compress) in transformers 5.x
+        if self.rope_parameters.get("full_attention", self.rope_parameters).get("rope_type") is None:
+            if (rope_scaling := raw_hparams.get("rope_scaling")) is not None:
+                if "rope_type" not in rope_scaling and (rope_type := rope_scaling.get("type")) is not None:
+                    rope_scaling["rope_type"] = rope_type
+                self.rope_parameters.update(**rope_scaling)
+
         self.block_count = self.hparams["num_hidden_layers"]
         if self.mtp_only:
             self.block_count += self.hparams.get("num_nextn_predict_layers", 0)