rcca: test default kv_cache_reuse option for pytorch multimodal (NVIDIA#5544)

StanleySun639 · dominicshanshan · commit 91e0362763af · 2025-07-10T02:20:57.000-07:00
Signed-off-by: Stanley Sun &lt;190317771+StanleySun639@users.noreply.github.com&gt;
diff --git a/tests/integration/defs/test_e2e.py b/tests/integration/defs/test_e2e.py
@@ -2043,7 +2043,6 @@ def test_ptp_quickstart_multimodal(llm_root, llm_venv, model_name, model_path,
         *accuracy_inputs[modality]["prompt"],
         "--media",
         *accuracy_inputs[modality]["media"],
-        "--disable_kv_cache_reuse",
     ]
     # NOTE: Qwen2-VL and Qwen2-5-VL model need larger max_num_tokens for video.
     if model_name in ["qwen2-vl-7b-instruct", "qwen2.5-vl-7b-instruct"

Original file line number	Diff line number	Diff line change
`@@ -2043,7 +2043,6 @@ def test_ptp_quickstart_multimodal(llm_root, llm_venv, model_name, model_path,`
`2043`	`2043`	`*accuracy_inputs[modality]["prompt"],`
`2044`	`2044`	`"--media",`
`2045`	`2045`	`*accuracy_inputs[modality]["media"],`
`2046`		`- "--disable_kv_cache_reuse",`
`2047`	`2046`	`]`
`2048`	`2047`	`# NOTE: Qwen2-VL and Qwen2-5-VL model need larger max_num_tokens for video.`
`2049`	`2048`	`if model_name in ["qwen2-vl-7b-instruct", "qwen2.5-vl-7b-instruct"`