Skip to content

GitLab

Unverified Commit 44b1fbe2 authored by fzyzcjy's avatar fzyzcjy Committed by GitHub
Browse files

Fix DeepSeek chunked prefill memory issue (#11149)

parent c0dbbdd1
Hide whitespace changes
Inline Side-by-side
Showing with 1 addition and 0 deletions
python/sglang/srt/models/deepseek_v2.py
View file @ 44b1fbe2
...@@ -1965,6 +1965,7 @@ class DeepseekV2AttentionMLA(nn.Module): ...@@ -1965,6 +1965,7 @@ class DeepseekV2AttentionMLA(nn.Module):
tmp_lse = torch.empty_like(accum_lse) tmp_lse = torch.empty_like(accum_lse)
merge_state_v2(output, lse, accum_output, accum_lse, tmp_output, tmp_lse) merge_state_v2(output, lse, accum_output, accum_lse, tmp_output, tmp_lse)
accum_output, accum_lse = tmp_output, tmp_lse accum_output, accum_lse = tmp_output, tmp_lse
del kv, k, v, output, lse, tmp_output, tmp_lse
return accum_output return accum_output
......
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