[Ascend]Support qwen3.5 (#18544)

This PR affects only the NPU. If any issues arise, please contact iforgetmyname.
2026-02-12 15:22:47 +08:00
parent 0305d12df2
commit 1edc69be08
3 changed files with 23 additions and 4 deletions
@@ -193,6 +193,15 @@ class ModelSlimConfig(QuantizationConfig):
    ):
        # adapted from vllm.model_executor.layers.quantization.utils.quant_utils.is_layer_skipped
        proj_name = prefix.split(".")[-1]
+        if not hasattr(self, "_quant_description_normalized"):
+            quant_description = {}
+            for prefix_, value in self.quant_description.items():
+                prefix_ = prefix_.replace("language_model.", "")
+                if "visual" in prefix_:
+                    prefix_ = prefix_.replace("model.", "")
+                quant_description[prefix_] = value
+            self.quant_description = quant_description
+            self._quant_description_normalized = True
        if proj_name in fused_mapping:
            shard_prefixes = [
                prefix.replace(proj_name, shard_proj_name)