[Perf][MRV2] Allow FULL decode graphs for one-token prompt tails (#58400)

Signed-off-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: Nicolò Lucchesi <nicolo.lucchesi@mistral.ai>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Co-authored-by: Yongye Zhu <zyy1102000@gmail.com>
This commit is contained in:
Nick Hill
2026-09-28 12:48:52 -07:00
committed by GitHub
co-authored by Nicolò Lucchesi Claude Opus 5.5 mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> Yongye Zhu
parent ef42093a36
commit e7c903609f
12 changed files with 144 additions and 42 deletions
+1
View File
@@ -96,6 +96,7 @@ def _make_batch(
num_computed_prefill_tokens_np=np.zeros(num_reqs, dtype=np.int32),
is_prefilling_np=np.zeros(num_reqs, dtype=np.bool_),
has_prefill=False,
decode_graph_eligible=True,
input_ids=torch.zeros(num_tokens, dtype=torch.int32, device=DEVICE),
positions=torch.arange(num_tokens, dtype=torch.int64, device=DEVICE),
is_padding=torch.zeros(num_tokens, dtype=torch.bool, device=DEVICE),