[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/model_executor/layers/moe/ep.py

@@ -28,38 +28,9 @@ except:
 
 import fastdeploy
 from fastdeploy.config import MoEPhase
+from fastdeploy.model_executor.layers.moe.moe import get_moe_scores
 from fastdeploy.utils import singleton
 
-try:
-    from fastdeploy.model_executor.ops.gpu import noaux_tc
-except:
-    logger.warning("import noaux_tc Failed!")
-
-
-def get_moe_scores(
-    gating_output: paddle.Tensor,
-    n_group,
-    topk_group,
-    top_k,
-    routed_scaling_factor,
-    e_score_correction_bias,
-) -> paddle.Tensor:
-    """
-    compute moe scores using e_score_correction_bias.
-    """
-    scores = paddle.nn.functional.sigmoid(gating_output)
-    assert e_score_correction_bias is not None, "e_score_correction_bias is none!"
-    scores_with_bias = scores + e_score_correction_bias
-    scores, topk_values, topk_idx = noaux_tc(
-        scores,
-        scores_with_bias,
-        n_group if n_group > 0 else 1,
-        topk_group if topk_group > 0 else 1,
-        top_k,
-        routed_scaling_factor,
-    )
-    return scores, topk_values, topk_idx
-
 
 @singleton
 class DeepEPEngine: