[Bugfix] Load checkpoint KV-cache scales in the Qwen3.5 family

Checkpoints that ship calibrated k_scale/v_scale tensors (e.g.
unsloth/Qwen3.8-27B-NVFP4, which carries one pair per full-attention
layer) name them ...self_attn.{k,v}_scale, but the runtime parameters
live on the inner attention layer (...self_attn.attn.{k,v}_scale).
Qwen3_5Model.load_weights hands the iterable straight to
AutoWeightsLoader, which performs no scale-name remapping, so the
scales are silently dropped and --kv-cache-dtype fp8 runs at scale 1.0
with only a warning_once that fires even on success (the MTP draft
layer legitimately has no scales).

Route scale-suffixed names through maybe_remap_kv_scale_name first,
mirroring gemma4/gpt_oss. No behavior change for checkpoints without
scales.

Measured on RTX 5090 / Qwen3.8-27B NVFP4 / fp8 KV / MTP k=3 (GSM8K
greedy, n=300): 92.3% at scale 1.0 vs 93.7% with scales loaded
(bf16-KV control 94.5%).

Fixes #54623

Co-Authored-By: Hinata <hinata@idihore.id>
Signed-off-by: MOVZX <movzx@yahoo.com>

vllm/model_executor/models/qwen3_5.py

@@ -61,6 +61,9 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,
 )
+from vllm.model_executor.model_loader.weight_utils import (
+    maybe_remap_kv_scale_name,
+)
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import MultiModalFeatureSpec
 from vllm.sequence import IntermediateTensors
@@ -296,8 +299,28 @@ class Qwen3_5Model(Qwen3NextModel):
                 n_shared_experts=1,
                 ckpt_prefix="mlp.shared_expert",
             )
+        # Checkpoints that ship calibrated KV-cache scales name them
+        # `...self_attn.{k,v}_scale`, but the runtime parameters live on the
+        # inner attention layer (`...self_attn.attn.{k,v}_scale`), and
+        # AutoWeightsLoader performs no remapping — without this the scales
+        # are silently dropped and fp8 KV runs at scale 1.0 (#54623).
+        params_dict = dict(self.named_parameters())
+
+        def _remap_kv_scale_names(
+            weights: Iterable[tuple[str, torch.Tensor]],
+        ) -> Iterable[tuple[str, torch.Tensor]]:
+            for name, weight in weights:
+                if name.endswith((".k_scale", ".v_scale", ".q_scale")):
+                    remapped = maybe_remap_kv_scale_name(name, params_dict)
+                    if remapped is None:
+                        continue
+                    name = remapped
+                yield name, weight
+
         loader = AutoWeightsLoader(self)
-        return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)
+        return loader.load_weights(
+            _remap_kv_scale_names(weights), mapper=self.hf_to_vllm_mapper
+        )
 
 
 class Qwen3_5ForCausalLMBase(