[Bugfix] DeepSeek Accuracy (#14476)

Signed-off-by: Lucas Wilkinson <[email protected]>

Author: LucasWilkinson

vllm/v1/attention/backends/mla/common.py

@@ -222,8 +222,7 @@
     Fp8LinearGenericOp, current_platform_fp8_dtype, is_fp8)
 from vllm.model_executor.layers.quantization.utils.quant_utils import (
     scaled_quantize)
-from vllm.model_executor.layers.rotary_embedding import (
-    DeepseekScalingRotaryEmbedding, RotaryEmbedding)
+from vllm.model_executor.layers.rotary_embedding import RotaryEmbedding
 from vllm.utils import cdiv, round_down
 
 try:
@@ -626,9 +625,12 @@ def __init__(
         self.qk_head_dim = qk_head_dim
         self.v_head_dim = v_head_dim
 
-        self.rotary_emb = rotary_emb
-        self.use_yarn_rope = isinstance(rotary_emb,
-                                        DeepseekScalingRotaryEmbedding)
+        # Hack for V1 for now to avoid torch library overhead (since we are
+        # already inside an attention custom op), pull out the forward
+        # method from the rotary embedding and call it directly
+        # TODO(lucas): we should probably find a cleaner way to do this
+        self.rotary_emb = rotary_emb._forward_method
+
         self.q_proj = q_proj
         self.kv_b_proj = kv_b_proj
         self.o_proj = o_proj