issues with LaMed-Phi3 inference

[lzl199704]

Cannot use the demo code for inference.

Transformers==4.46.2
Torch==2.4.0

[baifanxxx]

Hi,

Thank you for your attention, please try a lower version of transformers.

[YongchengYAO]

transformers==4.45 works for me

[Coisini-Glenda]

This problem occurred during the eval_caption test：

RuntimeError: Error(s) in loading state_dict for Phi3ForCausalLM:
size mismatch for model.embed_tokens.weight: copying a param with shape torch.Size([32015, 3072]) from checkpoint, the shape in current model is torch.Size([32064, 3072]).
size mismatch for lm_head.weight: copying a param with shape torch.Size([32015, 3072]) from checkpoint, the shape in current model is torch.Size([32064, 3072]).

[zhouhoo]

transformers==4.45 works for me

I got error as follow: Token indices sequence length is longer than the specified maximum sequence length for this model (2963 > 512). Running this sequence through the model will result in indexing errors 0%| | 0/64 [00:02<?, ?it/s] Traceback (most recent call last): File "/root/M3D/Bench/eval/eval_vqa.py", line 162, in <module> main() File "/root/M3D/Bench/eval/eval_vqa.py", line 114, in main generation = model.generate(image, input_id, max_new_tokens=args.max_new_tokens, File "/root/miniconda3/envs/py310/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context return func(*args, **kwargs) File "/root/miniconda3/envs/py310/lib/python3.10/site-packages/transformers/generation/utils.py", line 1804, in generate generation_config, model_kwargs = self._prepare_generation_config(generation_config, **kwargs) File "/root/miniconda3/envs/py310/lib/python3.10/site-packages/transformers/generation/utils.py", line 1355, in _prepare_generation_config model_kwargs = generation_config.update(**kwargs) AttributeError: 'Tensor' object has no attribute 'update'

it is version problem?