Garbled text generated when using qwen3 as inference in gen_ea_answer.py

[Rachelcoll]

Does anyone has successfully run the qwen3 evaluation script? When I ran EAGLE/eagle/evaluation/gen_ea_answer_qwen3.py, the output are meaningless text, and I think the base model forward() method of Qwen3Model in EAGLE/eagle/model/modeling_qwen3_kv.py may cause the error since the base_model inference is incorrect, does anyone encounter the same issue? thanks a lot for the help

[lixinqi7]

I got this error when using qwen3-8B.
File "/home/xxx/work/EAGLE/eagle/model/cnets.py", line 89, in repeat_kv
return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.