EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

CUDA-graph-compatible releasing and resuming KV cache and model weight memory (#2630)

Browse Source
This commit is contained in:
fzyzcjy
2025-01-14 03:38:51 +08:00
committed by GitHub
parent d08c77c434
commit 923f518337
12 changed files with 406 additions and 60 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
scripts/ci_install_dependency.sh
View File

@@ -12,8 +12,9 @@ bash "${SCRIPT_DIR}/killall_sglang.sh"
pip install --upgrade pip
pip install -e "python[all]" --find-links https://flashinfer.ai/whl/cu124/torch2.4/flashinfer/
# Force reinstall flashinfer
# Force reinstall flashinfer and torch_memory_saver
pip install flashinfer==0.1.6 --find-links ${FLASHINFER_REPO} --force-reinstall --no-deps
pip install torch_memory_saver --force-reinstall
pip install transformers==4.45.2 sentence_transformers accelerate peft