model: support dots.vlm1 model (#8778)

Co-authored-by: weishi <bushou@xiaohongshu.com>
Co-authored-by: Ezra-Yu <1105212286@qq.com>
Co-authored-by: Jianfei Wang <905787410@qq.com>
Co-authored-by: qianwu <wangjianfei@xiaohongshu.com>
This commit is contained in:
chenge@xiaohongshu.com
2025-09-12 17:38:38 +08:00
committed by GitHub
parent 6d40308905
commit 1b1701f1f7
11 changed files with 806 additions and 11 deletions

View File

@@ -216,6 +216,7 @@ class ModelConfig:
or "DeepseekV3ForCausalLMNextN" in self.hf_config.architectures
or "LongcatFlashForCausalLM" in self.hf_config.architectures
or "LongcatFlashForCausalLMNextN" in self.hf_config.architectures
or "DotsVLMForCausalLM" in self.hf_config.architectures
):
self.head_dim = 256
self.attention_arch = AttentionArch.MLA
@@ -734,6 +735,7 @@ multimodal_model_archs = [
"Phi4MMForCausalLM",
"VILAForConditionalGeneration",
"Step3VLForConditionalGeneration",
"DotsVLMForCausalLM",
]