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 9, 16:00 UTC
feat: Complete LoRA support for MiniMaxM2 Fixes #32736 (#32763)
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
david guan committed
Jan 24, 2026 at 20:48 UTC
bc0d291bfebf0979dbd6e8486de40aa95eb57490
2 files changed
+11
-3
docs/models/supported_models.md
+1
-1
@@ -422,7 +422,7 @@ th {
422
| `MiMoV2FlashForCausalLM` | MiMoV2Flash | `XiaomiMiMo/MiMo-V2-Flash`, etc. | ︎| ✅︎ |
423
| `MiniCPMForCausalLM` | MiniCPM | `openbmb/MiniCPM-2B-sft-bf16`, `openbmb/MiniCPM-2B-dpo-bf16`, `openbmb/MiniCPM-S-1B-sft`, etc. | ✅︎ | ✅︎ |
424
| `MiniCPM3ForCausalLM` | MiniCPM3 | `openbmb/MiniCPM3-4B`, etc. | ✅︎ | ✅︎ |
425
-| `MiniMaxM2ForCausalLM` | MiniMax-M2, MiniMax-M2.1 |`MiniMaxAI/MiniMax-M2`, etc. | | ✅︎ |
425
+| `MiniMaxM2ForCausalLM` | MiniMax-M2, MiniMax-M2.1 |`MiniMaxAI/MiniMax-M2`, etc. | ✅︎ | ✅︎ |
426
| `MistralForCausalLM` | Ministral-3, Mistral, Mistral-Instruct | `mistralai/Ministral-3-3B-Instruct-2512`, `mistralai/Mistral-7B-v0.1`, `mistralai/Mistral-7B-Instruct-v0.1`, etc. | ✅︎ | ✅︎ |
427
| `MistralLarge3ForCausalLM` | Mistral-Large-3-675B-Base-2512, Mistral-Large-3-675B-Instruct-2512 | `mistralai/Mistral-Large-3-675B-Base-2512`, `mistralai/Mistral-Large-3-675B-Instruct-2512`, etc. | ✅︎ | ✅︎ |
428
| `MixtralForCausalLM` | Mixtral-8x7B, Mixtral-8x7B-Instruct | `mistralai/Mixtral-8x7B-v0.1`, `mistralai/Mixtral-8x7B-Instruct-v0.1`, `mistral-community/Mixtral-8x22B-v0.1`, etc. | ✅︎ | ✅︎ |
vllm/model_executor/models/minimax_m2.py
+10
-2
@@ -59,7 +59,7 @@ from vllm.model_executor.model_loader.weight_utils import (
59
)
60
from vllm.sequence import IntermediateTensors
61
62
-from .interfaces import SupportsPP
62
+from .interfaces import SupportsLoRA, SupportsPP
63
from .utils import (
64
AutoWeightsLoader,
65
PPMissingLayer,
@@ -484,7 +484,15 @@ class MiniMaxM2Model(nn.Module):
484
return loaded_params
485
486
487
-class MiniMaxM2ForCausalLM(nn.Module, SupportsPP):
487
+class MiniMaxM2ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
488
+ packed_modules_mapping = {
489
+ "qkv_proj": [
490
+ "q_proj",
491
+ "k_proj",
492
+ "v_proj",
493
+ ],
494
+ }
495
+
496
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
497
super().__init__()
498
config = vllm_config.model_config.hf_config