Skip to content

Conversation

merrymercy
Copy link
Contributor

@merrymercy merrymercy commented Aug 13, 2024

with flashinfer >= 0.1.5, we can use a single workspace buffer for all wrappers.

@merrymercy merrymercy marked this pull request as ready for review August 13, 2024 11:15
@zhyncs
Copy link
Member

zhyncs commented Aug 14, 2024

@merrymercy merrymercy force-pushed the flashinfer-merge-buffer branch 2 times, most recently from 6322cde to 7383909 Compare August 15, 2024 01:02
@merrymercy merrymercy force-pushed the flashinfer-merge-buffer branch from 7383909 to 44ed280 Compare August 15, 2024 01:05
@merrymercy merrymercy enabled auto-merge (squash) August 15, 2024 01:12
@@ -27,7 +27,7 @@ def __init__(self):
# Runtime constants: others
self.num_continue_decode_steps = 10
self.retract_decode_steps = 20
self.flashinfer_workspace_size = 192 * 1024 * 1024
self.flashinfer_workspace_size = 384 * 1024 * 1024
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah, we truly need this large work space... otherwise the chunked prefill would probably crash.

@merrymercy merrymercy disabled auto-merge August 15, 2024 02:25
@merrymercy merrymercy merged commit 326df4b into main Aug 15, 2024
5 of 6 checks passed
@merrymercy merrymercy deleted the flashinfer-merge-buffer branch August 15, 2024 02:25
@zhyncs
Copy link
Member

zhyncs commented Aug 15, 2024

Why the UT failed?

timethink pushed a commit to timethink/sglang that referenced this pull request Mar 9, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants