[Model] Extend device-side mm normalization to Qwen3VL/Qwen3.5/Qwen4Next (#51289)

Signed-off-by: cjackal <44624812+cjackal@users.noreply.github.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
cjackal
2026-09-29 11:44:54 +08:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 07f1ba2ec6
commit 491f44adfa
11 changed files with 87 additions and 25 deletions
@@ -21,6 +21,7 @@ from vllm.model_executor.layers.fused_moe import (
FusedMoEFactory,
GateLinear,
)
from vllm.model_executor.layers.fusion.mm_input_norm import build_mm_input_norm
from vllm.model_executor.layers.linear import ReplicatedLinear
from vllm.model_executor.layers.logits_processor import LogitsProcessor
from vllm.model_executor.layers.mamba.gdn.qwen_gdn_linear_attn import (
@@ -496,6 +497,7 @@ class InternS2MobiusForConditionalGeneration(InternS2PreviewForConditionalGenera
config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=quant_config,
input_norm=build_mm_input_norm(self.model_config),
prefix=maybe_prefix(prefix, "visual"),
)