Update rope_scaling to rope_parameters in preparation for Transformers v5 (#28542)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

vllm/model_executor/models/seed_oss.py

@@ -54,6 +54,7 @@ from vllm.model_executor.model_loader.weight_utils import (
     maybe_remap_kv_scale_name,
 )
 from vllm.sequence import IntermediateTensors
+from vllm.transformers_utils.config import set_default_rope_theta
 
 from .interfaces import SupportsLoRA, SupportsPP
 from .utils import (
@@ -112,11 +113,10 @@ class SeedOssAttention(nn.Module):
         num_heads: int,
         num_kv_heads: int,
         head_dim: int,
+        rope_parameters: dict,
         max_position: int = 4096 * 32,
-        rope_theta: float = 10000,
         cache_config: CacheConfig | None = None,
         quant_config: QuantizationConfig | None = None,
-        rope_scaling: tuple | None = None,
         prefix: str = "",
         attn_type: str = AttentionType.DECODER,
     ) -> None:
@@ -140,7 +140,6 @@ class SeedOssAttention(nn.Module):
         self.q_size = self.num_heads * self.head_dim
         self.kv_size = self.num_kv_heads * self.head_dim
         self.scaling = self.head_dim**-0.5
-        self.rope_theta = rope_theta
 
         self.qkv_proj = QKVParallelLinear(
             hidden_size,
@@ -163,8 +162,7 @@ class SeedOssAttention(nn.Module):
             self.head_dim,
             rotary_dim=self.head_dim,
             max_position=max_position,
-            base=self.rope_theta,
-            rope_scaling=rope_scaling,
+            rope_parameters=rope_parameters,
         )
         self.attn = Attention(
             self.num_heads,
@@ -200,9 +198,7 @@ class SeedOssDecoderLayer(nn.Module):
     ) -> None:
         super().__init__()
         self.hidden_size = config.hidden_size
-        # Requires transformers > 4.32.0
-        rope_theta = getattr(config, "rope_theta", 1000000)
-        rope_scaling = getattr(config, "rope_scaling", None)
+        set_default_rope_theta(config, default_theta=1000000)
 
         # By default, SeedOss uses causal attention as it is a
         # decoder-only model.
@@ -219,10 +215,9 @@ class SeedOssDecoderLayer(nn.Module):
             max_position=config.max_position_embeddings,
             num_kv_heads=config.num_key_value_heads,
             head_dim=config.head_dim,
-            rope_theta=rope_theta,
             cache_config=cache_config,
             quant_config=quant_config,
-            rope_scaling=rope_scaling,
+            rope_parameters=config.rope_parameters,
             prefix=f"{prefix}.self_attn",
             attn_type=attn_type,
         )