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 5, 06:00 UTC
[CI][Metrics] Fix local_cache_hit assertion after prompt tokens metrics updates (#39709)
Signed-off-by: ZhanqiuHu <zhu@redhat.com>
zhanqiuhu committed
Apr 13, 2026 at 11:16 UTC
10d9872d3aee629a47374347c523964eed8cad00
1 file changed
+4
-4
tests/v1/kv_connector/nixl_integration/test_multi_connector_edge_cases.py
+4
-4
@@ -341,8 +341,8 @@ def test_full_decode_gpu_cache_hit_metrics():
341
print(f"FULL CACHE HIT: {P} tokens, cached={cached}, nixl={expected_nixl}")
342
print(f" metrics delta: {d}, nixl_bytes_delta={n1 - n0}")
343
assert len(proxy_text) > 0, "proxy returned empty response"
344
- assert d["local_cache_hit"] == cached, (
345
- f"expected local_cache_hit={cached}, got {d['local_cache_hit']}"
344
+ assert d["local_cache_hit"] == cached - 1, (
345
+ f"expected local_cache_hit={cached - 1}, got {d['local_cache_hit']}"
346
)
347
assert d["external_kv_transfer"] == expected_nixl, (
348
f"expected external_kv_transfer={expected_nixl}, "
@@ -383,8 +383,8 @@ def test_partial_decode_gpu_cache_hit_metrics():
383
f"expected external_kv_transfer={expected_nixl}, "
384
f"got {d['external_kv_transfer']}"
385
)
386
- assert d["local_cache_hit"] == cached, (
387
- f"expected local_cache_hit={cached}, got {d['local_cache_hit']}"
386
+ assert d["local_cache_hit"] == cached - 1, (
387
+ f"expected local_cache_hit={cached - 1}, got {d['local_cache_hit']}"
388
)
389
assert d["local_compute"] == 1, (
390
f"expected local_compute=1 (recomputed last token), got {d['local_compute']}"