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 4, 08:45 UTC
Test nemotron nano-v2 and nemotron nano-v3 separately, disable super-omni redundant tests (#41616)
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
Netanel Haber committed
May 4, 2026 at 16:31 UTC
8decbfa02c9bbc1699b2136ecc72b1ef30c438a0
1 file changed
+10
-41
tests/models/registry.py
+10
-41
@@ -1141,30 +1141,17 @@ _MULTIMODAL_EXAMPLE_MODELS = {
1141
"NemotronH_Nano_VL_V2": _HfExamplesInfo(
1142
"nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16",
1143
max_model_len=4096,
1144
- # NemotronH layers are constructed via `hybrid_override_pattern`:
1144
+ # NemotronH layers are constructed via `hybrid_override_pattern`
1145
use_original_num_layers=True,
1146
hf_overrides={
1147
- "vision_config": PretrainedConfig(
1148
- args={
1149
- "min_num_patches": 1, # Trigger image dynamic res
1150
- "max_num_patches": 12,
1151
- "model": "vit_huge_patch16_224",
1152
- },
1153
- # Trigger conv3d:
1154
- video_temporal_patch_size=2,
1155
- ),
1156
- "text_config": {
1157
- "num_hidden_layers": 2,
1158
- "hybrid_override_pattern": "M*",
1159
- },
1147
+ "text_config": {"num_hidden_layers": 2, "hybrid_override_pattern": "M*"},
1148
},
1149
trust_remote_code=True,
1150
),
1163
- # NemotronH_Nano_Omni_Reasoning_V3 is an alias for NemotronH_Nano_VL_V2
1164
- # Use the same registry test as NemotronH_Nano_VL_V2 above
1151
"NemotronH_Nano_Omni_Reasoning_V3": _HfExamplesInfo(
1166
- "nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16",
1152
+ "nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16",
1153
max_model_len=4096,
1154
+ # NemotronH layers are constructed via `hybrid_override_pattern`
1155
use_original_num_layers=True,
1156
hf_overrides={
1157
"vision_config": PretrainedConfig(
@@ -1174,35 +1161,17 @@ _MULTIMODAL_EXAMPLE_MODELS = {
1161
"model": "vit_huge_patch16_224",
1162
},
1163
video_temporal_patch_size=2,
1164
+ # TODO(nhaber): This is `true` in the official `config.json`,
1165
+ # but this causes a processor exception in the tests due to a known bug
1166
+ # with mixed-resolution video when `true`. To be resolved.
1167
+ video_maintain_aspect_ratio=False,
1168
),
1178
- "text_config": {
1179
- "num_hidden_layers": 2,
1180
- "hybrid_override_pattern": "M*",
1181
- },
1169
+ "text_config": {"num_hidden_layers": 2, "hybrid_override_pattern": "M*"},
1170
},
1171
trust_remote_code=True,
1172
),
1185
- # NemotronH_Super_Omni_Reasoning_V3 is an alias for NemotronH_Nano_VL_V2 as well
1186
- # Use the same registry test as NemotronH_Nano_VL_V2 above
1173
"NemotronH_Super_Omni_Reasoning_V3": _HfExamplesInfo(
1188
- "nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16",
1189
- max_model_len=4096,
1190
- use_original_num_layers=True,
1191
- hf_overrides={
1192
- "vision_config": PretrainedConfig(
1193
- args={
1194
- "min_num_patches": 1,
1195
- "max_num_patches": 12,
1196
- "model": "vit_huge_patch16_224",
1197
- },
1198
- video_temporal_patch_size=2,
1199
- ),
1200
- "text_config": {
1201
- "num_hidden_layers": 2,
1202
- "hybrid_override_pattern": "M*",
1203
- },
1204
- },
1205
- trust_remote_code=True,
1174
+ "nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16", is_available_online=False
1175
),
1176
"OpenCUAForConditionalGeneration": _HfExamplesInfo(
1177
"xlangai/OpenCUA-7B",