[Test] Test multiple attn backend for chunked prefill. (#4023)

2024-04-13 01:56:57 +09:00
parent 7fd3949a0b
commit 36729bac13
4 changed files with 13 additions and 23 deletions
--- a/tests/basic_correctness/test_chunked_prefill.py
+++ b/tests/basic_correctness/test_chunked_prefill.py
@@ -33,10 +33,6 @@ def test_models(
    enforce_eager: bool,
    tensor_parallel_size: int,
 ) -> None:
-    if (tensor_parallel_size == 2 and chunked_prefill_token_size != 16
-            and not enforce_eager):
-        pytest.skip(f"Skip {chunked_prefill_token_size=} and {enforce_eager=} "
-                    "for high TP to save testing time.")
    max_num_seqs = min(chunked_prefill_token_size, 256)
    enable_chunked_prefill = False
    max_num_batched_tokens = None