fix: kvcache evict workflow

2026-04-14 15:46:36 +08:00
parent 663ca9c5b9
commit eaf574cd4e
4 changed files with 257 additions and 59 deletions
--- a/configs/glm5-nvfp4-8xb300.yaml
+++ b/configs/glm5-nvfp4-8xb300.yaml
@@ -14,11 +14,12 @@ model:
 hardware:
  type: 8xb300
  hbm_bytes: 1900.0e9          # KV budget after FP4 weights (~372 GB)
+  dram_bytes: 1.5e12             # ~1.5 TB usable CPU DRAM per node

 cluster:
-  num_instances: 32
+  num_instances: 8
  meta_store:
-    ttl_seconds: 120.0
+    ttl_seconds: 300.0
  router:
    mode: prefix_affinity
    prefix_k: 8