nvidia/Eagle2-9B · Version Crash for Qwen2 from Transformers

Thanks for great work!

I would like to suggest removing "configuration_qwen2.py" and "modeling_qwen2.py" and just replacing them with "from transformers.models.qwen2.configuration_qwen2 import Qwen2Config" and "from transformers import Qwen2ForCausalLM"

It must be leading Transformer version error as I was also facing them, but I resolved it when I remove their files and manually modify the code like below

From "configuration_eagle_chat.py" Line 13

from transformers.models.qwen2.configuration_qwen2 import Qwen2Config

From "modeling_eagle_chat.py" Line 51-52

        # assert version_cmp(transformers.__version__, '4.37.2', 'ge')
        # assert version_cmp(transformers.__version__, '4.39.2', 'le')

From "modeling_eagle_chat.py" Line 23

from transformers import Qwen2ForCausalLM

From "modeling_eagle_chat.py" Line 441-449 [We should remove return_dict argument]

        outputs = self.language_model.generate(
            inputs_embeds=input_embeds,
            attention_mask=attention_mask,
            generation_config=generation_config,
            output_hidden_states=output_hidden_states,
            use_cache=True,
            **generate_kwargs,
        )