[CP]Glm45 air 2.2 (#4073)

* [Feature] Support zai-org/GLM-4.5-Air BF16 model (#3928)

* support glm45_air

* [Feature] GLM-45-AIR Support Mix Quantization(Dense wfp8afp8 and wint8 triton_moe_backend) (#4051)

* check

* fix v1 load for mix and wint8

* check --quantizations 'None'

* check

* support RL rollout

* check v1 loader

* check glm rollout_model, change wfp8afp8 per_token_cast_to_fp8 to native impl

* check rollout moe gate begin layer_id

* check rollout e_score_correction_bias

* delete infer_to_train_mapping={}

* code check

fastdeploy/engine/args_utils.py

@@ -41,6 +41,7 @@ from fastdeploy.utils import (
     DeprecatedOptionWarning,
     FlexibleArgumentParser,
     is_port_available,
+    parse_quantization,
 )
 
 
@@ -138,7 +139,7 @@ class EngineArgs:
     """
     dynamic load weight strategy
     """
-    quantization: str = None
+    quantization: Optional[Dict[str, Any]] = None
     guided_decoding_backend: str = "off"
     """
     Guided decoding backend.
@@ -550,7 +551,7 @@ class EngineArgs:
         )
         model_group.add_argument(
             "--quantization",
-            type=str,
+            type=parse_quantization,
             default=EngineArgs.quantization,
             help="Quantization name for the model, currentlly support "
             "'wint8', 'wint4',"