[Quant] Fix use_mla TypeError and support loading pure-sparsity Compressed Tensors configs (#12711)

2025-02-04 02:27:11 -05:00
parent bb392af434
commit 4896d0c2dd
2 changed files with 8 additions and 2 deletions
--- a/vllm/config.py
+++ b/vllm/config.py
@@ -1000,8 +1000,9 @@ class ModelConfig:
        # have fp8 for both weights and activations.
        if self.quantization == "compressed-tensors":
            quant_config = self._parse_quant_hf_config()
-            for group_name, cfg in quant_config.get("config_groups",
-                                                    ("", {})).items():
+            for group_name, cfg in quant_config.get("config_groups", {
+                    "": {}
+            }).items():
                act_cfg = cfg.get("input_activations", {})
                act_type = None if act_cfg is None else act_cfg.get("type", "")
                w_cfg = cfg.get("weights", {})