Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix the possible bug of decode out of memory #36

Merged
merged 15 commits into from
Jan 19, 2024
Merged

Fix the possible bug of decode out of memory #36

Changes from 1 commit
Commits
Show all changes
15 commits
Select commit Hold shift + click to select a range
c98e251
organize infer_batch
hnyls2002 Jan 18, 2024
21a9383
update return_normalized_logprob when batch changes
hnyls2002 Jan 18, 2024
da43356
add suspend for decode
hnyls2002 Jan 18, 2024
8a1b1b4
fixed suspend
hnyls2002 Jan 18, 2024
ae2d162
remove old
hnyls2002 Jan 18, 2024
e3d7a81
evict less
hnyls2002 Jan 18, 2024
5b0fe48
Merge branch 'main' into decode-fix
hnyls2002 Jan 18, 2024
9413710
clear
hnyls2002 Jan 18, 2024
402b3c4
add robust test
hnyls2002 Jan 19, 2024
082b052
rename
hnyls2002 Jan 19, 2024
8686079
evict more for avoiding overhead
hnyls2002 Jan 19, 2024
bdebcfb
fixed json fsm init
hnyls2002 Jan 19, 2024
a92ffff
dynamic estimate ratio
hnyls2002 Jan 19, 2024
571a07c
format
merrymercy Jan 19, 2024
c49edea
fix test cases
merrymercy Jan 19, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
remove old
  • Loading branch information
@hnyls2002
hnyls2002 committed Jan 18, 2024
commit ae2d1625de883479b550dc48d1a2021f32181443
15 changes: 3 additions & 12 deletions python/sglang/srt/managers/router/infer_batch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -246,11 +246,6 @@ def suspend_for_decode(self):
reverse=True,
)

for i in range(len(self.reqs)):
print(
f"input: {len(self.reqs[i].input_ids)}, output: {len(self.reqs[i].output_ids)}"
)

suspended_reqs = []
seq_lens_np = self.seq_lens.cpu().numpy()
req_pool_indices_np = self.req_pool_indices.cpu().numpy()
Expand Down Expand Up @@ -289,13 +284,9 @@ def prepare_for_decode(self, input_ids=None):
self.out_cache_loc = self.token_to_kv_pool.alloc(bs)

if self.out_cache_loc is None:
self.tree_cache.evict(bs, self.token_to_kv_pool.free)
self.out_cache_loc = self.token_to_kv_pool.alloc(bs)

if self.out_cache_loc is None:
print("Decode out of memory.")
self.tree_cache.pretty_print()
exit()
print("Decode out of memory. This should nerver happen.")
self.tree_cache.pretty_print()
exit()

self.out_cache_cont_start = None
self.out_cache_cont_end = None
Expand Down