A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 12:15 UTC
[TPU][V1][CI] Update regression test baseline for v6 CI (#17064)
Signed-off-by: NickLucche <nlucches@redhat.com>
Nicolò Lucchesi committed
Apr 30, 2025 at 13:03 UTC
a7d5b016bd0a882767dc1c3cc1537dc8c93a2ea7
1 file changed
+7
-8
tests/v1/tpu/test_perf.py
+7
-8
@@ -59,17 +59,16 @@ TEST_PARAMS = [
59
prefix_len=500,
60
decode_len=50,
61
62
- # (This is the active CI/CD instance)
62
# commit id: ccb246776d93ef105904a8ec015b3587240a1183
64
- # tpu: v5lite (vllm CI/CD)
65
- expected_avg_time=1.4,
66
- err_tol=0.30,
63
+ # tpu: v5lite (old vllm CI/CD)
64
+ # expected_avg_time=1.4,
65
+ # err_tol=0.30,
66
68
- # (TODO: There is no v6e in CI/CD currently)
67
+ # (This is the active CI/CD instance)
68
# commit id: ccb246776d93ef105904a8ec015b3587240a1183
70
- # tpu: v6e
71
- # expected_avg_time=1.5,
72
- # err_tol=0.20,
69
+ # tpu: v6e (current vllm CI/CD)
70
+ expected_avg_time=1.7, # measured with VLLM_XLA_CACHE_PATH=
71
+ err_tol=0.20,
72
),
73
]
74