[MRV2] Revert explicit Triton JIT warmup migration (#56654)

Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
Co-authored-by: OpenAI Codex <codex@openai.com>
This commit is contained in:
Woosuk Kwon
2026-09-13 04:07:56 +00:00
committed by GitHub
co-authored by OpenAI Codex
parent c711f740b7
commit 2671fedfc7
6 changed files with 96 additions and 227 deletions
@@ -43,7 +43,7 @@ def _run_prepare(
num_scheduled_tokens=np.array([4], dtype=np.int32),
positions=torch.tensor(target_positions, dtype=torch.int64, device=device),
query_start_loc=torch.tensor([0, 4], dtype=torch.int32, device=device),
idx_mapping=torch.tensor([2], dtype=torch.int64, device=device),
idx_mapping=torch.tensor([2], dtype=torch.int32, device=device),
)
query_slot_mapping = torch.full(
(max_num_tokens,), -2, dtype=torch.int64, device=device
@@ -157,7 +157,6 @@ def test_mm_support_configured_after_model_load(monkeypatch):
speculator.device = device
speculator.max_num_tokens = 4
speculator.max_num_reqs = 2
speculator.num_speculative_steps = 1
speculator.hidden_size = 3
speculator.dtype = torch.float32
speculator.draft_model_config = draft_model_config