From 2974f8be59a5cf0d26e23f84515fa70ed4dc9885 Mon Sep 17 00:00:00 2001 From: tzhouam Date: Sat, 3 Oct 2026 21:20:08 +0800 Subject: [PATCH] test: use the vLLM 0.31 model-length property in graph fixtures Signed-off-by: tzhouam --- tests/worker_v2/test_decode_query_bound_gpu.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tests/worker_v2/test_decode_query_bound_gpu.py b/tests/worker_v2/test_decode_query_bound_gpu.py index 114e9f343..9089fb76b 100644 --- a/tests/worker_v2/test_decode_query_bound_gpu.py +++ b/tests/worker_v2/test_decode_query_bound_gpu.py @@ -54,7 +54,7 @@ def test_full_decode_graph_matches_fa3_aot_schedule(graph_mode, tmp_path): kvconfig = KVCacheConfig(num_blocks=1024, kv_cache_tensors=[], kv_cache_groups=[KVCacheGroupSpec(["attn"], spec)]) state = OmniModelState.__new__(OmniModelState) state.vllm_config = c - state.max_model_len = 2048 + state.model_config = c.model_config state.supports_mm_inputs = False with set_current_vllm_config(c): group = AttentionGroup(FlashAttentionBackend, ["attn"], spec, 0)