Skip to content

[Bug fix] Fix cudagraph when use ep. #3130

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

Merged
RichardWooSJTU merged 3 commits into from
Aug 4, 2025
Merged

[Bug fix] Fix cudagraph when use ep. #3130

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
68a630a
fix cudagraph when use ep
Wanglongzhi2001 Aug 1, 2025
836cc05
fix typo
Wanglongzhi2001 Aug 1, 2025
dc6d657
reduce full length to adapt large bsz such 128/256
Wanglongzhi2001 Aug 4, 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
6 changes: 6 additions & 0 deletions fastdeploy/worker/gpu_model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -525,6 +525,12 @@ def _dummy_prefill_inputs(self, num_tokens: int, batch_size: int, expected_decod
num_tokens // batch_size,
self.parallel_config.max_model_len - max_dec_len,
)

# NOTE(wanglongzhi): When the full length is too large, DeepEP's buffer size will not be enough to cause the result to appear nan.
# TODO(wanglongzhi): Figure out the accurate buffer size of DeepEP.
if self.fd_config.parallel_config.enable_expert_parallel:
full_length = min(full_length, 32)

input_length = int(full_length * self.cache_config.kv_cache_ratio)
block_num = (
input_length + self.cache_config.block_size - 1
Expand Down
Loading