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, 13:45 UTC
[Model] update for mimo v25 (#41029)
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com> Signed-off-by: Isotr0py <Isotr0py@outlook.com> Co-authored-by: Isotr0py <Isotr0py@outlook.com> Co-authored-by: Copilot <copilot@github.com>
Jiangyun Zhu committed
Apr 28, 2026 at 12:52 UTC
7a1eb8ac2ec4ea69338c51dc7afd4b15010abfa8
6 files changed
+10
-8
docs/models/supported_models.md
+1
-1
@@ -439,7 +439,7 @@ th {
439
| `Mamba2ForCausalLM` | Mamba2 | `mistralai/Mamba-Codestral-7B-v0.1`, etc. | | ✅︎ |
440
| `MiMoForCausalLM` | MiMo | `XiaomiMiMo/MiMo-7B-RL`, etc. | ✅︎ | ✅︎ |
441
| `MiMoV2FlashForCausalLM` | MiMoV2Flash | `XiaomiMiMo/MiMo-V2-Flash`, etc. | | ✅︎ |
442
-| `MiMoV2ProForCausalLM` | MiMoV2Pro | `XiaomiMiMo/MiMo-V2.5-Pro`, etc. | | ✅︎ |
442
+| `MiMoV2ForCausalLM` | MiMoV2Pro | `XiaomiMiMo/MiMo-V2.5-Pro`, etc. | | ✅︎ |
443
| `MiniCPMForCausalLM` | MiniCPM | `openbmb/MiniCPM-2B-sft-bf16`, `openbmb/MiniCPM-2B-dpo-bf16`, `openbmb/MiniCPM-S-1B-sft`, etc. | ✅︎ | ✅︎ |
444
| `MiniCPM3ForCausalLM` | MiniCPM3 | `openbmb/MiniCPM3-4B`, etc. | ✅︎ | ✅︎ |
445
| `MiniMaxForCausalLM` | MiniMax-Text | `MiniMaxAI/MiniMax-Text-01-hf`, etc. | | |
tests/models/registry.py
+2
-2
@@ -594,8 +594,8 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
594
"MiMoV2FlashForCausalLM": _HfExamplesInfo(
595
"XiaomiMiMo/MiMo-V2-Flash", trust_remote_code=True
596
),
597
- "MiMoV2ProForCausalLM": _HfExamplesInfo(
598
- "XiaomiMiMo/MiMo-V2.5-Pro", trust_remote_code=True, is_available_online=False
597
+ "MiMoV2ForCausalLM": _HfExamplesInfo(
598
+ "XiaomiMiMo/MiMo-V2.5-Pro", trust_remote_code=True
599
),
600
"Dots1ForCausalLM": _HfExamplesInfo("rednote-hilab/dots.llm1.inst"),
601
}
vllm/config/speculative.py
+2
-2
@@ -339,7 +339,7 @@ class SpeculativeConfig:
339
)
340
341
if (arch := hf_config.architectures[0]) in (
342
- "MiMoV2ProForCausalLM",
342
+ "MiMoV2ForCausalLM",
343
"MiMoV2OmniForCausalLM",
344
):
345
from vllm.model_executor.models.mimo_v2_mtp import (
@@ -347,7 +347,7 @@ class SpeculativeConfig:
347
)
348
349
mtp_arch_maps = {
350
- "MiMoV2ProForCausalLM": "MiMoV2MTPModel",
350
+ "MiMoV2ForCausalLM": "MiMoV2MTPModel",
351
"MiMoV2OmniForCausalLM": "MiMoV2OmniMTPModel",
352
}
353
vllm/model_executor/models/mimo_v2.py
+1
-1
@@ -733,7 +733,7 @@ class MiMoV2FlashForCausalLM(nn.Module, SupportsPP, MixtureOfExperts):
733
return loader.load_weights(weights)
734
735
736
-class MiMoV2ProForCausalLM(MiMoV2FlashForCausalLM):
736
+class MiMoV2ForCausalLM(MiMoV2FlashForCausalLM):
737
packed_modules_mapping = {
738
"qkv_proj": ["qkv_proj"],
739
"gate_up_proj": ["gate_proj", "up_proj"],
vllm/model_executor/models/registry.py
+1
-1
@@ -172,7 +172,7 @@ _TEXT_GENERATION_MODELS = {
172
"MPTForCausalLM": ("mpt", "MPTForCausalLM"),
173
"MiMoForCausalLM": ("mimo", "MiMoForCausalLM"),
174
"MiMoV2FlashForCausalLM": ("mimo_v2", "MiMoV2FlashForCausalLM"),
175
- "MiMoV2ProForCausalLM": ("mimo_v2", "MiMoV2ProForCausalLM"),
175
+ "MiMoV2ForCausalLM": ("mimo_v2", "MiMoV2ForCausalLM"),
176
"NemotronForCausalLM": ("nemotron", "NemotronForCausalLM"),
177
"NemotronHForCausalLM": ("nemotron_h", "NemotronHForCausalLM"),
178
"NemotronHPuzzleForCausalLM": ("nemotron_h", "NemotronHForCausalLM"),
vllm/transformers_utils/model_arch_config_convertor.py
+3
-1
@@ -458,6 +458,8 @@ def _strip_mimo_v2_attention_chunk_size(
458
459
class MimoV2ModelArchConfigConvertor(ModelArchConfigConvertorBase):
460
def __init__(self, hf_config: PretrainedConfig, hf_text_config: PretrainedConfig):
461
+ if getattr(hf_config, "vision_config", None):
462
+ hf_config.architectures = ["MiMoV2OmniForCausalLM"]
463
super().__init__(hf_config, hf_text_config)
464
_strip_mimo_v2_attention_chunk_size(hf_config, hf_text_config)
465
@@ -542,7 +544,7 @@ MODEL_ARCH_CONFIG_CONVERTORS = {
544
"qwen3_next_mtp": Qwen3NextMTPModelArchConfigConvertor,
545
"qwen3_5_mtp": Qwen3_5MTPModelArchConfigConvertor,
546
"mimo_mtp": MimoMTPModelArchConfigConvertor,
545
- "mimo_v2_pro": MimoV2ModelArchConfigConvertor,
547
+ "mimo_v2": MimoV2ModelArchConfigConvertor,
548
"mimo_v2_flash": MimoV2ModelArchConfigConvertor,
549
"mimo_v2_mtp": MimoV2MTPModelArchConfigConvertor,
550
"mimo_v2_omni_mtp": MimoV2MTPModelArchConfigConvertor,