[V1] [Hybrid] Support Minimax-Text-01 in V1 (#22151)

Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>
2025-08-09 08:08:48 +02:00
parent 3157aebb63
commit 6ade99eafa
5 changed files with 234 additions and 42 deletions
--- a/vllm/model_executor/layers/mamba/mamba_utils.py
+++ b/vllm/model_executor/layers/mamba/mamba_utils.py
@@ -5,6 +5,17 @@ from vllm.distributed import divide

 class MambaStateShapeCalculator:

+    @classmethod
+    def linear_attention_state_shape(
+        cls,
+        num_heads: int,
+        tp_size: int,
+        head_dim: int,
+    ) -> tuple[tuple[int, int, int], ...]:
+
+        state_shape = (num_heads // tp_size, head_dim, head_dim)
+        return (state_shape, )
+
    @classmethod
    def mamba1_state_shape(
        cls,