Resolve CUDA UT failure with transformers 5.0.0rc #1315
Description
test report: AutoRound Cuda Unit Test
use transformers=5.0.0rc3
- test_2_3bits @xin3he lm_eval api
- test_alg_ext @xin3he lm_eval api
- test_asym @xin3he lm_eval api
- test_auto_round_format @xin3he lm_eval api
- test_auto_scheme @xin3he lm_eval api
- test_conv1d @xin3he transformers api( from transformers.modeling_utils import no_init_weights)
- test_diffusion @n1ck-guo transformers api (ImportError: cannot import name 'MT5Tokenizer' from 'transformers')
- test_exllamav2_backend @xin3he lm_eval api
- test_export @n1ck-guotransformers api (ImportError: cannot import name 'MT5Tokenizer' from 'transformers')
- test_fp8_input @xin3he lm_eval api
- test_get_block_name @n1ck-guo transformers api (ImportError: cannot import name 'MT5Tokenizer' from 'transformers')
- test_gguf @Kaihui-intel cuda ut no token.model
- test_main_func @xin3he lm_eval api
- test_marlin_backend @xin3he lm_eval api
- test_mix_bits @n1ck-guo transformers api
- test_moe_model @xin3he
- test_multiple_card @n1ck-guo
- test_mxfp_nvfp @xin3he
- test_qbits @n1ck-guo
- test_scheme @n1ck-guo
- test_sglang @xin3he fix: Refactor register_image_processor to use kwarg instead of positional arg sgl-project/sglang#17685
- test_torch_backend @xin3he lm_eval
- test_transformers @xin3he [BUG] cannot import name 'no_init_weights' from 'transformers.modeling_utils' ModelCloud/GPTQModel#2328
- test_triton_backend @xin3he lm_eval api
- test_vllm @xin3he [CI Failure]: Transformers Nightly Models Test vllm-project/vllm#31181