EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

[Platform] format platform to make it more clear (#610)

Browse Source 
Platform should only contain the function that based from vllm. This PR
move the unrelated function to the right place to make platform more
clear.

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-04-30 09:03:10 +08:00
committed by GitHub
parent b917361ca5
commit 95e7aa4736
3 changed files with 17 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm_ascend/patch/__init__.py
View File

@@ -62,7 +62,7 @@
# inside of the repo, and needs a common interface to destroy them, this patch add the interface of destroy
# platform owned `CoordinatorGroup` to make sure all the CoordinateGroup can be properly destroyed
# How：
# Call platform method `destroy_platform_model_parallel` to destroy all the `CoordinateGroup`
# Call `vllm_ascend.distributed.parallel_state method `destroy_platform_model_parallel` to destroy all the `CoordinateGroup`
# Related PR (if no, explain why): no related PR, we want add this ability into vllm
# Future Plan:
# Remove those patch when vllm merged them
@@ -73,7 +73,7 @@
# call to the `stateless_init_torch_distributed_process_group`, to enable other platform which may support
# stateless process group initialize method
# How：
# Call platform method `platform_has_backend_register` to judge if there is a stateless process group initialize
# rewrite stateless_init_torch_distributed_process_group to judge if there is a stateless process group initialize
# method and call platform method `platform_register_backend` to initialize them
# Related PR (if no, explain why): no related PR, we want add this ability into vllm
# Future Plan: