Unverified Commit 7c347259 authored by Zilin Zhu's avatar Zilin Zhu Committed by GitHub
Browse files

[RL] allow weight updation with dp attention enabled (#6311)

parent 669caa0a
...@@ -893,8 +893,8 @@ class TokenizerManager: ...@@ -893,8 +893,8 @@ class TokenizerManager:
) -> Tuple[bool, str]: ) -> Tuple[bool, str]:
self.auto_create_handle_loop() self.auto_create_handle_loop()
assert ( assert (
self.server_args.dp_size == 1 self.server_args.dp_size == 1 or self.server_args.enable_dp_attention
), "dp_size must be 1 for update weights from distributed" ), "dp_size must be 1 or dp attention must be enabled for update weights from distributed"
# This means that weight sync # This means that weight sync
# cannot run while requests are in progress. # cannot run while requests are in progress.
...@@ -909,8 +909,8 @@ class TokenizerManager: ...@@ -909,8 +909,8 @@ class TokenizerManager:
) -> Tuple[bool, str]: ) -> Tuple[bool, str]:
self.auto_create_handle_loop() self.auto_create_handle_loop()
assert ( assert (
self.server_args.dp_size == 1 self.server_args.dp_size == 1 or self.server_args.enable_dp_attention
), "dp_size must be 1 for update weights from distributed" ), "dp_size must be 1 or dp attention must be enabled for update weights from tensor"
# This means that weight sync # This means that weight sync
# cannot run while requests are in progress. # cannot run while requests are in progress.
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment