[NemotronH] PP support (#16172)

Signed-off-by: Roi Koren <roik@nvidia.com>
2025-12-31 11:16:15 +08:00
parent c0fc7a89e7
commit 47a660d5b9
2 changed files with 94 additions and 35 deletions
@@ -11,6 +11,12 @@ class TestNvidiaNemotronNanoV2BF16(GSM8KMixin, DefaultServerBase):
    other_args = ["--max-mamba-cache-size", "256"]


+class TestNvidiaNemotronNanoV2BF16PP(GSM8KMixin, DefaultServerBase):
+    model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2"
+    gsm8k_accuracy_thres = 0.87
+    other_args = ["--max-mamba-cache-size", "256", "--pp-size", "2"]
+
+
 class TestNvidiaNemotronNanoV2FP8(GSM8KMixin, DefaultServerBase):
    gsm8k_accuracy_thres = 0.87
    model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2-FP8"