[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/utils.py

@@ -18,6 +18,7 @@ import argparse
 import asyncio
 import codecs
 import importlib
+import json
 import logging
 import os
 import random
@@ -757,6 +758,18 @@ class StatefulSemaphore:
         }
 
 
+def parse_quantization(value: str):
+    """
+    Parse a JSON string into a dictionary.
+    """
+    try:
+        return json.loads(value)
+    except ValueError:
+        if value is None or value.lower() == "none":
+            return None
+        return {"quantization": value}
+
+
 # 日志使用全局访问点（兼容原有使用方式）
 def get_logger(name, file_name=None, without_formater=False, print_to_console=False):
     """全局函数包装器，保持向后兼容"""