[bugfix] fix inductor cache on max_position_embeddings (#15436)

Signed-off-by: youkaichao <youkaichao@gmail.com>
2025-03-25 22:05:39 +08:00 · 2025-03-25 22:05:39 +08:00 · d0cfec7ab9
commit d0cfec7ab9
parent a608160027
1 changed files with 3 additions and 0 deletions
--- a/vllm/config.py
+++ b/vllm/config.py
@ -221,6 +221,9 @@ class ModelConfig:
        factors.append(self.trust_remote_code)
        factors.append(self.rope_scaling)
        factors.append(self.rope_theta)
+        # rope cos/sin cache depends on the max_position_embeddings
+        factors.append(
+            getattr(self.hf_config, "max_position_embeddings", "None"))
        return hashlib.sha256(str(factors).encode()).hexdigest()

    def __init__(