[GLM-4.7] GLM-Lite Supoort (#43031)
* draft
* update
* Delete test_text.py
* update
* ignore
* Delete model.safetensors.index_30.json
* update for test
* Update tokenization_auto.py
* update test
* Update test_modeling_glm4_moe_lite.py
* update
* update
* using GradientCheckpointingLayer
* 2
* 3
* remove first_k_dense_replace
* 1
