[Feature] Option to save model weights to CPU when memory saver mode is enabled (#10873)

Co-authored-by: molocule <34072934+molocule@users.noreply.github.com>
2025-10-03 04:48:19 -04:00
parent 34151f173b
commit 8c57490210
7 changed files with 78 additions and 22 deletions
--- a/python/pyproject_other.toml
+++ b/python/pyproject_other.toml
@@ -110,7 +110,7 @@ srt_hpu = ["sglang[runtime_common]"]
 openai = ["openai==1.99.1", "tiktoken"]
 anthropic = ["anthropic>=0.20.0"]
 litellm = ["litellm>=1.0.0"]
-torch_memory_saver = ["torch_memory_saver==0.0.8"]
+torch_memory_saver = ["torch_memory_saver==0.0.9rc1"]
 decord = ["decord"]
 test = [
    "accelerate",