router: fix Dense context window 98K→262K (matches actual n_ctx)
This commit is contained in:
+1
-1
@@ -27,7 +27,7 @@ GPU_MAX_CONCURRENT = {
|
|||||||
# Context window sizes (tokens) — used for compaction signals
|
# Context window sizes (tokens) — used for compaction signals
|
||||||
GPU_CONTEXT = {
|
GPU_CONTEXT = {
|
||||||
"qwen3.6-35B-A3B": 131072,
|
"qwen3.6-35B-A3B": 131072,
|
||||||
"qwen3.6-27B-code": 98304,
|
"qwen3.6-27B-code": 262144,
|
||||||
"gemma-4-12b": 262144, # 262K context (gemma-4-12b)
|
"gemma-4-12b": 262144, # 262K context (gemma-4-12b)
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user