[Model] Add MiMo-V2.5 support (#40967)

Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Signed-off-by: Isotr0py <Isotr0py@outlook.com>
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: Jee Jee Li <pandaleefree@gmail.com>
Co-authored-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: zjy0516 <zhujiangyun@inferact.ai>
Co-authored-by: yasong <yasong.wang@inferact.ai>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Copilot <copilot@github.com>
This commit is contained in:
Isotr0py
2026-04-27 13:26:51 +00:00
committed by GitHub
co-authored by Jee Jee Li zjy0516 zjy0516 yasong Claude Opus 4.6 Copilot
parent f8ac0c7cf0
commit c245d35ff4
16 changed files with 4737 additions and 5 deletions
+1 -1
View File
@@ -521,6 +521,7 @@ class ModelConfig:
if dict_overrides:
self._apply_dict_overrides(hf_config, dict_overrides)
self.hf_text_config = get_hf_text_config(self.hf_config)
self.model_arch_config = self.get_model_arch_config()
self.attention_chunk_size = getattr(
self.hf_text_config, "attention_chunk_size", None
)
@@ -528,7 +529,6 @@ class ModelConfig:
self.hf_image_processor_config = get_hf_image_processor_config(
self.model, hf_token=self.hf_token, revision=self.revision
)
self.model_arch_config = self.get_model_arch_config()
architectures = self.architectures
registry = self.registry
+43
View File
@@ -34,6 +34,7 @@ logger = init_logger(__name__)
MTPModelTypes = Literal[
"deepseek_mtp",
"mimo_mtp",
"mimo_v2_mtp",
"glm4_moe_mtp",
"glm4_moe_lite_mtp",
"glm_ocr_mtp",
@@ -332,6 +333,48 @@ class SpeculativeConfig:
}
)
if (arch := hf_config.architectures[0]) in (
"MiMoV2ProForCausalLM",
"MiMoV2OmniForCausalLM",
):
from vllm.model_executor.models.mimo_v2_mtp import (
_MIMO_V2_PRO_NUM_MTP_LAYERS,
)
mtp_arch_maps = {
"MiMoV2ProForCausalLM": "MiMoV2MTPModel",
"MiMoV2OmniForCausalLM": "MiMoV2OmniMTPModel",
}
hf_config.model_type = "mimo_v2_mtp"
# vLLM currently supports only the first MiMo-V2 MTP layer.
n_predict = _MIMO_V2_PRO_NUM_MTP_LAYERS
hf_config.update(
{
"num_hidden_layers": 0,
"n_predict": n_predict,
"num_nextn_predict_layers": n_predict,
"architectures": [mtp_arch_maps[arch]],
}
)
if hf_config.architectures[0] == "MiMoV2FlashForCausalLM":
from vllm.model_executor.models.mimo_v2_mtp import (
_MIMO_V2_FLASH_NUM_MTP_LAYERS,
)
hf_config.model_type = "mimo_v2_mtp"
# vLLM currently supports only the first MiMo-V2 MTP layer.
n_predict = _MIMO_V2_FLASH_NUM_MTP_LAYERS
hf_config.update(
{
"num_hidden_layers": 0,
"n_predict": n_predict,
"num_nextn_predict_layers": n_predict,
"architectures": ["MiMoV2MTPModel"],
}
)
if hf_config.architectures[0] == "Glm4MoeForCausalLM":
hf_config.model_type = "glm4_moe_mtp"
n_predict = getattr(hf_config, "num_nextn_predict_layers", None)