Fix conversion of fused qkvz & FP8 for Qwen3-Next

conversion/base.py

@@ -620,6 +620,9 @@ class ModelBase:
     def _transform_fp8_scale(self, name: str, scale: Tensor) -> Tensor:
         return scale
 
+    def _map_fp8_weight_names(self, name: str) -> tuple[str, ...]:
+        return (self.map_tensor_name(name),)
+
     def _prepare_fp8_e4m3_tensors(self):
         if self._fp8_as_q8:
             return
@@ -686,10 +689,10 @@ class ModelBase:
                     assert isinstance(entries, list)
                     cast(list[tuple[int, float]], entries).append((expert_id, float(input_scale[0])))
             else:
-                new_name = self.map_tensor_name(weight_name)
-                scale_tensors[new_name.replace(".weight", ".scale")] = scale.numpy()
-                if input_scale is not None:
-                    input_scale_tensors[new_name.replace(".weight", ".input_scale")] = input_scale.numpy()
+                for new_name in self._map_fp8_weight_names(weight_name):
+                    scale_tensors[new_name.replace(".weight", ".scale")] = scale.numpy()
+                    if input_scale is not None:
+                        input_scale_tensors[new_name.replace(".weight", ".input_scale")] = input_scale.numpy()
 
         for name in consumed:
             self.model_tensors.pop(name, None)

conversion/qwen.py

@@ -373,6 +373,16 @@ class _QwenMtpMixin:
 class Qwen3NextModel(_QwenMtpMixin, Qwen2MoeModel):
     model_arch = gguf.MODEL_ARCH.QWEN3NEXT
 
+    def _map_fp8_weight_names(self, name: str) -> tuple[str, ...]:
+        if name.endswith(".linear_attn.in_proj_qkvz.weight"):
+            bid = next(int(part) for part in self.map_tensor_name(name).split(".") if part.isdecimal())
+            # Both split projections keep the original per-tensor scales.
+            return (
+                self.format_tensor_name(gguf.MODEL_TENSOR.ATTN_QKV, bid),
+                self.format_tensor_name(gguf.MODEL_TENSOR.ATTN_GATE, bid),
+            )
+        return super()._map_fp8_weight_names(name)
+
     def set_gguf_parameters(self):
         super().set_gguf_parameters()
         self.gguf_writer.add_ssm_conv_kernel(self.hparams["linear_conv_kernel_dim"])

src/models/qwen3next.cpp

@@ -324,11 +324,11 @@ std::pair<ggml_tensor *, ggml_tensor *> llama_model_qwen3next::graph::build_qkvz
 
     if (model.layers[il].wqkv) {
         // optimized path
-        ggml_tensor * qkv_mixed = build_lora_mm(model.layers[il].wqkv, input);
+        ggml_tensor * qkv_mixed = build_lora_mm(model.layers[il].wqkv, input, model.layers[il].wqkv_s);
         qkv_mixed = ggml_reshape_3d(ctx0, qkv_mixed, qkv_mixed->ne[0], n_seq_tokens, n_seqs);
         cb(qkv_mixed, "linear_attn_qkv_mixed", il);
 
-        ggml_tensor * z = build_lora_mm(model.layers[il].wqkv_gate, input);
+        ggml_tensor * z = build_lora_mm(model.layers[il].wqkv_gate, input, model.layers[il].wqkv_gate_s);
         cb(z, "z", il);
 
         return { qkv_mixed, z };