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 7, 05:15 UTC
[Chore] Correct MTP models test registry ordering (#36115)
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Isotr0py committed
Mar 5, 2026 at 16:55 UTC
21eb2c3372fb6447ef36bee44ff7af79a330ffec
1 file changed
+32
-29
tests/models/registry.py
+32
-29
@@ -421,11 +421,6 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
421
"Olmo2ForCausalLM": _HfExamplesInfo("allenai/OLMo-2-0425-1B"),
422
"Olmo3ForCausalLM": _HfExamplesInfo("allenai/Olmo-3-7B-Instruct"),
423
"OlmoeForCausalLM": _HfExamplesInfo("allenai/OLMoE-1B-7B-0924-Instruct"),
424
- "OpenPanguMTPModel": _HfExamplesInfo(
425
- "FreedomIntelligence/openPangu-Ultra-MoE-718B-V1.1",
426
- trust_remote_code=True,
427
- is_available_online=False,
428
- ),
424
"OPTForCausalLM": _HfExamplesInfo(
425
"facebook/opt-125m", {"1b": "facebook/opt-iml-max-1.3b"}
426
),
@@ -1018,14 +1013,6 @@ _MULTIMODAL_EXAMPLE_MODELS = {
1013
"Qwen/Qwen3.5-35B-A3B",
1014
max_model_len=4096,
1015
),
1021
- "Qwen3_5MTP": _HfExamplesInfo(
1022
- "Qwen/Qwen3.5-0.8B",
1023
- speculative_model="Qwen/Qwen3.5-0.8B",
1024
- ),
1025
- "Qwen3_5MoeMTP": _HfExamplesInfo(
1026
- "Qwen/Qwen3.5-35B-A3B",
1027
- speculative_model="Qwen/Qwen3.5-35B-A3B",
1028
- ),
1016
"Qwen3OmniMoeForConditionalGeneration": _HfExamplesInfo(
1017
"Qwen/Qwen3-Omni-30B-A3B-Instruct",
1018
max_model_len=4096,
@@ -1093,6 +1080,7 @@ _MULTIMODAL_EXAMPLE_MODELS = {
1080
1081
1082
_SPECULATIVE_DECODING_EXAMPLE_MODELS = {
1083
+ # [Medusa]
1084
"MedusaModel": _HfExamplesInfo(
1085
"JackFram/llama-68m", speculative_model="abhigoyal/vllm-medusa-llama-68m-random"
1086
),
@@ -1102,11 +1090,7 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
1090
# "JackFram/llama-160m",
1091
# speculative_model="ibm-ai-platform/llama-160m-accelerator"
1092
# ),
1105
- "DeepSeekMTPModel": _HfExamplesInfo(
1106
- "luccafong/deepseek_mtp_main_random",
1107
- speculative_model="luccafong/deepseek_mtp_draft_random",
1108
- trust_remote_code=True,
1109
- ),
1093
+ # [Eagle]
1094
"EagleDeepSeekMTPModel": _HfExamplesInfo(
1095
"eagle618/deepseek-v3-random",
1096
speculative_model="eagle618/eagle-deepseek-v3-random",
@@ -1152,6 +1136,20 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
1136
speculative_method="eagle",
1137
tokenizer="openbmb/MiniCPM-2B-sft-bf16",
1138
),
1139
+ "Eagle3Qwen2_5vlForCausalLM": _HfExamplesInfo(
1140
+ "Qwen/Qwen2.5-VL-7B-Instruct",
1141
+ speculative_model="Rayzl/qwen2.5-vl-7b-eagle3-sgl",
1142
+ ),
1143
+ "Eagle3Qwen3vlForCausalLM": _HfExamplesInfo(
1144
+ "Qwen/Qwen3-VL-8B-Instruct",
1145
+ speculative_model="taobao-mnn/Qwen3-VL-8B-Instruct-Eagle3",
1146
+ ),
1147
+ # [MTP]
1148
+ "DeepSeekMTPModel": _HfExamplesInfo(
1149
+ "luccafong/deepseek_mtp_main_random",
1150
+ speculative_model="luccafong/deepseek_mtp_draft_random",
1151
+ trust_remote_code=True,
1152
+ ),
1153
"ErnieMTPModel": _HfExamplesInfo(
1154
"baidu/ERNIE-4.5-21B-A3B-PT",
1155
trust_remote_code=True,
@@ -1191,17 +1189,27 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
1189
trust_remote_code=True,
1190
speculative_model="XiaomiMiMo/MiMo-7B-RL",
1191
),
1194
- "Eagle3Qwen2_5vlForCausalLM": _HfExamplesInfo(
1195
- "Qwen/Qwen2.5-VL-7B-Instruct",
1196
- speculative_model="Rayzl/qwen2.5-vl-7b-eagle3-sgl",
1192
+ "NemotronHMTPModel": _HfExamplesInfo(
1193
+ "nvidia/Nemotron-Super-Placeholder",
1194
+ speculative_model="nvidia/Nemotron-Super-Placeholder",
1195
+ is_available_online=False,
1196
),
1198
- "Eagle3Qwen3vlForCausalLM": _HfExamplesInfo(
1199
- "Qwen/Qwen3-VL-8B-Instruct",
1200
- speculative_model="taobao-mnn/Qwen3-VL-8B-Instruct-Eagle3",
1197
+ "OpenPanguMTPModel": _HfExamplesInfo(
1198
+ "FreedomIntelligence/openPangu-Ultra-MoE-718B-V1.1",
1199
+ trust_remote_code=True,
1200
+ is_available_online=False,
1201
),
1202
"Qwen3NextMTP": _HfExamplesInfo(
1203
"Qwen/Qwen3-Next-80B-A3B-Instruct", min_transformers_version="4.56.3"
1204
),
1205
+ "Qwen3_5MTP": _HfExamplesInfo(
1206
+ "Qwen/Qwen3.5-0.8B",
1207
+ speculative_model="Qwen/Qwen3.5-0.8B",
1208
+ ),
1209
+ "Qwen3_5MoeMTP": _HfExamplesInfo(
1210
+ "Qwen/Qwen3.5-35B-A3B",
1211
+ speculative_model="Qwen/Qwen3.5-35B-A3B",
1212
+ ),
1213
"Step3p5MTP": _HfExamplesInfo(
1214
"stepfun-ai/Step-3.5-Flash",
1215
speculative_model="stepfun-ai/Step-3.5-Flash",
@@ -1212,11 +1220,6 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
1220
},
1221
is_available_online=False,
1222
),
1215
- "NemotronHMTPModel": _HfExamplesInfo(
1216
- "nvidia/Nemotron-Super-Placeholder",
1217
- speculative_model="nvidia/Nemotron-Super-Placeholder",
1218
- is_available_online=False,
1219
- ),
1223
}
1224
1225
_TRANSFORMERS_BACKEND_MODELS = {