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

Fix min_p sampling crash when using flashinfer backend #3207

Merged
merged 5 commits into from
Feb 2, 2025
Merged

Fix min_p sampling crash when using flashinfer backend #3207

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
a3660e3
Fix min_p sampling crash when using flashinfer backend
zifeitong Jan 29, 2025
6e0c1b1
Merge branch 'main' into fix_min_p
zifeitong Jan 31, 2025
3f61c4a
Merge branch 'main' into fix_min_p
zifeitong Feb 1, 2025
18e3d04
Merge branch 'main' into fix_min_p
zhaochenyang20 Feb 1, 2025
b16bd53
Merge branch 'main' into fix_min_p
zhaochenyang20 Feb 2, 2025
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
8 changes: 4 additions & 4 deletions python/sglang/srt/layers/sampler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -85,7 +85,7 @@ def forward(
if sampling_info.need_min_p_sampling:
probs = top_k_renorm_prob(probs, sampling_info.top_ks)
probs = top_p_renorm_prob(probs, sampling_info.top_ps)
batch_next_token_ids, success = min_p_sampling_from_probs(
batch_next_token_ids = min_p_sampling_from_probs(
probs, uniform_samples, sampling_info.min_ps
)
else:
Expand All @@ -97,9 +97,9 @@ def forward(
filter_apply_order="joint",
)

if self.use_nan_detectioin and not torch.all(success):
logger.warning("Detected errors during sampling!")
batch_next_token_ids = torch.zeros_like(batch_next_token_ids)
if self.use_nan_detectioin and not torch.all(success):
logger.warning("Detected errors during sampling!")
batch_next_token_ids = torch.zeros_like(batch_next_token_ids)

elif global_server_args_dict["sampling_backend"] == "pytorch":
# A slower fallback implementation with torch native operations.
Expand Down
Loading