apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

fix cuda graph capture failure in CI test (#7094)

Browse Source
This commit is contained in:
huicongyao
2026-03-31 11:05:51 +08:00
committed by GitHub
parent daa95244f7
commit dd2aa10ed4
1 changed files with 9 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
custom_ops/gpu_ops/speculate_decoding/verify_draft_tokens.cu
+9 -2
View File
@@ -252,9 +252,16 @@ __global__ void verify_draft_tokens(
break;
}
// Accept-all override (debug/warmup)
// Accept-all override (debug/warmup/CUDA graph capture)
if (accept_all) {
if (ctx.emit_token(i, ctx.step_input_ids_now[i + 1])) break;
int64_t token = ctx.step_input_ids_now[i + 1];
// During dummy run (accept_all), replace EOS tokens with a safe
// non-EOS value to prevent stop_flags being set, which would cause
// CUDA graph capture failure due to token count mismatch.
if (is_in_end(token, end_tokens, end_length)) {
token = 5;
}
if (ctx.emit_token(i, token)) break;
continue;
}