Changed to correct architecture name for vllm

Added num stages tuning

Changed to autotune triton for 48G GPU deployment

Removed assertion

Removed wrong edge case assertion

Corrected device assignment error

Corrected device assignment error

Corrected device assignment error

Safe cache check

Removed GenerationMixing

Hardwired nsa attention

Added attn_imp in config

Flattened repo

Merge branch 'main' of https://huggingface.co/Maxtimer97/GLM2NSA

Relative import first

Put relative import first for Huggingface

Added modeling files

Upload tokenizer

Upload ChatGLMForConditionalGeneration

initial commit