support absorb for prefill long context

2025-09-10 23:34:35 +00:00 · 2025-02-25 08:52:02 +00:00 · 2025-02-25 08:52:02 +00:00 · f4c198bd42
commit f4c198bd42
parent e9b1216a9a
8 changed files with 93 additions and 33 deletions
--- a/ktransformers/optimize/optimize_rules/DeepSeek-V3-Chat.yaml
+++ b/ktransformers/optimize/optimize_rules/DeepSeek-V3-Chat.yaml
@ -60,6 +60,7 @@
    kwargs:
      generate_device: "cuda"
      prefill_device: "cuda"
+      absorb_for_prefill: False # change this to True to enable long context(prefill may slower).
 - match:
    name: "^model$"
  replace: