diff --git a/router/router.py b/router/router.py index 6e9cfb8..c1a239e 100644 --- a/router/router.py +++ b/router/router.py @@ -27,7 +27,7 @@ GPU_MAX_CONCURRENT = { # Context window sizes (tokens) — used for compaction signals GPU_CONTEXT = { "qwen3.6-35B-A3B": 262144, - "qwen3.6-27B-code": 262144, + "qwen3.6-27B-code": 196608, "qwen3.5-9b-vlm": 262144, }